server: skip duplicate response on CancelledError#1153
Merged
ihrpr merged 9 commits intomodelcontextprotocol:mainfrom Jul 25, 2025
Merged
server: skip duplicate response on CancelledError#1153ihrpr merged 9 commits intomodelcontextprotocol:mainfrom
ihrpr merged 9 commits intomodelcontextprotocol:mainfrom
Conversation
When a tool call is cancelled, RequestResponder.cancel() sends an error response. Previously, _handle_request would try to send another error response, causing "Request already responded to" assertion. This fix catches CancelledError and returns early, preventing the duplicate response. Fixes modelcontextprotocol#1152
ihrpr
requested changes
Jul 21, 2025
Contributor
ihrpr
left a comment
ihrpr
left a comment
There was a problem hiding this comment.
Thanks for woking on the fix! Left some comments
tests/server/test_cancel_handling.py
Outdated
Contributor
There was a problem hiding this comment.
There is a lot of mocking happening. I'd suggest to use test_integration or leverage create_connected_server_and_client_session and test with CancelledNotification
Contributor
Author
There was a problem hiding this comment.
Fixed, thanks for the directions!
- Use anyio.get_cancelled_exc_class() instead of asyncio.CancelledError for proper backend compatibility (supports both asyncio and trio) - Rewrote cancel handling tests to use integration testing pattern with create_connected_server_and_client_session instead of mocking - Tests now properly send CancelledNotification messages through real client-server communication - Added proper type annotations and fixed type issues in tests Addresses review feedback about using anyio idioms and integration testing. Reported-by: @ihrpr
lukacf
force-pushed
the
fix-cancelled-error-response
branch
from
89075b0 to
a5a112a
Compare
ihrpr
requested changes
Jul 22, 2025
Contributor
ihrpr
left a comment
ihrpr
left a comment
There was a problem hiding this comment.
Thanks for the quick turnaround, I think we can remove test_cancelled_request_no_double_response test. wdyt?
tests/server/test_cancel_handling.py
Outdated
|
|
||
|
|
||
| @pytest.mark.anyio | ||
| async def test_cancelled_request_no_double_response(): |
Contributor
There was a problem hiding this comment.
why do we need this test? it doesn't look like it's testing much and test_server_remains_functional_after_cancel looks complete to cover the fix?
ihrpr
approved these changes
Jul 25, 2025
Contributor
ihrpr
left a comment
ihrpr
left a comment
There was a problem hiding this comment.
Thank you
(I removed the test mentioned in comments as test_server_remains_functional_after_cancel already covers the change)
rbehal
pushed a commit
to gumloop/gumloop-mcp
that referenced
this pull request
Sep 25, 2025
* Add regression test for stateless request memory cleanup (modelcontextprotocol#1140) * Implement RFC9728 - Support WWW-Authenticate header by MCP client (modelcontextprotocol#1071) * Add streamable HTTP starlette example to Python SDK docs (modelcontextprotocol#1111) * fix markdown error in README in main (modelcontextprotocol#1147) * README - replace code snippets with examples - add lowlevel to snippets (modelcontextprotocol#1150) * README - replace code snippets with examples - streamable http (modelcontextprotocol#1155) * chore: don't allow users to create issues outside the templates (modelcontextprotocol#1163) * Tests(cli): Add coverage for helper functions (modelcontextprotocol#635) * Docs: Update CallToolResult parsing in README (modelcontextprotocol#812) Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> * docs: add pre-commit install guide on CONTRIBUTING.md (modelcontextprotocol#995) Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> * fix flaky fix-test_streamablehttp_client_resumption test (modelcontextprotocol#1166) * README - replace code snippets with examples -- auth examples (modelcontextprotocol#1164) * Support falling back to OIDC metadata for auth (modelcontextprotocol#1061) * Add CODEOWNERS file for sdk (modelcontextprotocol#1169) * fix flaky test test_88_random_error (modelcontextprotocol#1171) * Make sure `RequestId` is not coerced as `int` (modelcontextprotocol#1178) * Fix: Replace threading.Lock with anyio.Lock for Ray deployment compatibility (modelcontextprotocol#1151) * fix: fix OAuth flow request object handling (modelcontextprotocol#1174) * update codeowners group (modelcontextprotocol#1191) * fix: perform auth server metadata discovery fallbacks on any 4xx (modelcontextprotocol#1193) * server: skip duplicate response on CancelledError (modelcontextprotocol#1153) Co-authored-by: ihrpr <inna@anthropic.com> * Unpack settings in FastMCP (modelcontextprotocol#1198) * chore: Remove unused prompt_manager.py file (modelcontextprotocol#1229) Co-authored-by: Tapan Chugh <tapanc@cs.washington.edu> * Improved supported for ProtectedResourceMetadata (modelcontextprotocol#1235) Co-authored-by: Paul Carleton <paulcarletonjr@gmail.com> * chore: Remove unused variable notification_options (modelcontextprotocol#1238) * Improve README around the Context object (modelcontextprotocol#1203) * fix: allow to pass `list[str]` to `token_endpoint_auth_signing_alg_values_supported` (modelcontextprotocol#1226) * Remove strict validation on `response_modes_supported` member of `OAuthMetadata` (modelcontextprotocol#1243) * Add pyright strict mode on the whole project (modelcontextprotocol#1254) * Consistent casing for default headers Accept and Content-Type (modelcontextprotocol#1263) * Update dependencies and fix type issues (modelcontextprotocol#1268) Co-authored-by: Marcelo Trylesinski <marcelotryle@gmail.com> * fix: prevent async generator cleanup errors in StreamableHTTP transport (modelcontextprotocol#1271) Co-authored-by: David Soria Parra <167242713+dsp-ant@users.noreply.github.com> * chore: uncomment .idea/ in .gitignore (modelcontextprotocol#1287) Co-authored-by: Claude <noreply@anthropic.com> * docs: clarify streamable_http_path configuration when mounting servers (modelcontextprotocol#1172) * feat: Add CORS configuration for browser-based MCP clients (modelcontextprotocol#1059) Co-authored-by: Marcelo Trylesinski <marcelotryle@gmail.com> Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> * Added Audio to FastMCP (modelcontextprotocol#1130) * fix: avoid uncessary retries in OAuth authenticated requests (modelcontextprotocol#1206) Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> * Add PATHEXT to default STDIO env vars in windows (modelcontextprotocol#1256) * fix: error too many values to unpack (expected 2) (modelcontextprotocol#1279) Signed-off-by: San Nguyen <vinhsannguyen91@gmail.com> Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * SDK Parity: Avoid Parsing Server Response for non-JsonRPCMessage Requests (modelcontextprotocol#1290) * types: Setting default value for method: Literal (modelcontextprotocol#1292) * changes structured temperature to not deadly (modelcontextprotocol#1328) * Update simple-resource example to use non-deprecated read_resource return type (modelcontextprotocol#1331) Co-authored-by: Claude <noreply@anthropic.com> * docs: Update README to include link to API docs for modelcontextprotocol#1329 (modelcontextprotocol#1330) * Allow ping requests before initialization (modelcontextprotocol#1312) * Python lint: Ruff rules for pylint and code complexity (modelcontextprotocol#525) * Fix context injection for resources and prompts (modelcontextprotocol#1336) * fix(fastmcp): propagate mimeType in resource template list (modelcontextprotocol#1186) Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * fix: allow elicitations accepted without content (modelcontextprotocol#1285) Co-authored-by: Olivier Schiavo <olivier.schiavo@wengo.com> * Use --frozen in pre-commit config (modelcontextprotocol#1375) * Return HTTP 403 for invalid Origin headers (modelcontextprotocol#1353) * Add test for ProtectedResourceMetadataParsing (modelcontextprotocol#1236) Co-authored-by: Paul Carleton <paulcarletonjr@gmail.com> Co-authored-by: Marcelo Trylesinski <marcelotryle@gmail.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * Fastmcp logging progress example (modelcontextprotocol#1270) Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * feat: add paginated list decorators for prompts, resources, and tools (modelcontextprotocol#1286) Co-authored-by: Claude <noreply@anthropic.com> * Remove "unconditionally" from conditional description (modelcontextprotocol#1289) * Use streamable-http consistently in examples (modelcontextprotocol#1389) * feat: Add SDK support for SEP-1034 default values in elicitation schemas (modelcontextprotocol#1337) Co-authored-by: Tapan Chugh <tapanc@cs.washington.edu> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * Implementation of SEP 973 - Additional metadata + icons support (modelcontextprotocol#1357) * Merge upstream/main with custom filtering --------- Signed-off-by: San Nguyen <vinhsannguyen91@gmail.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> Co-authored-by: yurikunash <143175350+yurikunash@users.noreply.github.com> Co-authored-by: Pamela Fox <pamela.fox@gmail.com> Co-authored-by: Inna Harper <inna.hrpr@gmail.com> Co-authored-by: Marcelo Trylesinski <marcelotryle@gmail.com> Co-authored-by: Ian Davenport <49379192+davenpi@users.noreply.github.com> Co-authored-by: Dagang Wei <functicons@gmail.com> Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> Co-authored-by: Stanley Law <stanleylkal@gmail.com> Co-authored-by: Luca Chang <131398524+LucaButBoring@users.noreply.github.com> Co-authored-by: leweng <leweng@nvidia.com> Co-authored-by: Clare Liguori <liguori@amazon.com> Co-authored-by: lukacf <luka@peltarion.com> Co-authored-by: ihrpr <inna@anthropic.com> Co-authored-by: Tapan Chugh <chugh.tapan@gmail.com> Co-authored-by: Tapan Chugh <tapanc@cs.washington.edu> Co-authored-by: Yann Jouanin <4557670+yannj-fr@users.noreply.github.com> Co-authored-by: Paul Carleton <paulcarletonjr@gmail.com> Co-authored-by: Sreenath Somarajapuram <somarajapuram@gmail.com> Co-authored-by: Omer Korner <omerkorner@gmail.com> Co-authored-by: joesavage-silabs <159480754+joesavage-silabs@users.noreply.github.com> Co-authored-by: Gregory L <gregory.linford@mistral.ai> Co-authored-by: David Soria Parra <167242713+dsp-ant@users.noreply.github.com> Co-authored-by: Moustapha Ebnou <155577789+mous222@users.noreply.github.com> Co-authored-by: Max Isbey <224885523+maxisbey@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Jerome <jerome@anthropic.com> Co-authored-by: xavier <84836280+dragonier23@users.noreply.github.com> Co-authored-by: keurcien <keurcien.luu@gmail.com> Co-authored-by: Tim Esler <tim.esler@gmail.com> Co-authored-by: San Nguyen <22189661+sandangel@users.noreply.github.com> Co-authored-by: Justin Wang <89049861+justin-yi-wang@users.noreply.github.com> Co-authored-by: jess <jessachandler@gmail.com> Co-authored-by: Peter Alexander <pja@anthropic.com> Co-authored-by: Reid Geyer <12072650+reidg44@users.noreply.github.com> Co-authored-by: Eleftheria Stein-Kousathana <eleftheria.kousathana@gmail.com> Co-authored-by: Christian Clauss <cclauss@me.com> Co-authored-by: pchoudhury22 <pchoudhury22@apple.com> Co-authored-by: owengo <owengo@users.noreply.github.com> Co-authored-by: Olivier Schiavo <olivier.schiavo@wengo.com> Co-authored-by: Steve Billings <billings.steve@gmail.com> Co-authored-by: Mike Salvatore <mike.s.salvatore@gmail.com>
rbehal
added a commit
to gumloop/gumloop-mcp
that referenced
this pull request
Dec 10, 2025
* Add regression test for stateless request memory cleanup (modelcontextprotocol#1140) * Implement RFC9728 - Support WWW-Authenticate header by MCP client (modelcontextprotocol#1071) * Add streamable HTTP starlette example to Python SDK docs (modelcontextprotocol#1111) * fix markdown error in README in main (modelcontextprotocol#1147) * README - replace code snippets with examples - add lowlevel to snippets (modelcontextprotocol#1150) * README - replace code snippets with examples - streamable http (modelcontextprotocol#1155) * chore: don't allow users to create issues outside the templates (modelcontextprotocol#1163) * Tests(cli): Add coverage for helper functions (modelcontextprotocol#635) * Docs: Update CallToolResult parsing in README (modelcontextprotocol#812) Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> * docs: add pre-commit install guide on CONTRIBUTING.md (modelcontextprotocol#995) Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> * fix flaky fix-test_streamablehttp_client_resumption test (modelcontextprotocol#1166) * README - replace code snippets with examples -- auth examples (modelcontextprotocol#1164) * Support falling back to OIDC metadata for auth (modelcontextprotocol#1061) * Add CODEOWNERS file for sdk (modelcontextprotocol#1169) * fix flaky test test_88_random_error (modelcontextprotocol#1171) * Make sure `RequestId` is not coerced as `int` (modelcontextprotocol#1178) * Fix: Replace threading.Lock with anyio.Lock for Ray deployment compatibility (modelcontextprotocol#1151) * fix: fix OAuth flow request object handling (modelcontextprotocol#1174) * update codeowners group (modelcontextprotocol#1191) * fix: perform auth server metadata discovery fallbacks on any 4xx (modelcontextprotocol#1193) * server: skip duplicate response on CancelledError (modelcontextprotocol#1153) Co-authored-by: ihrpr <inna@anthropic.com> * Unpack settings in FastMCP (modelcontextprotocol#1198) * chore: Remove unused prompt_manager.py file (modelcontextprotocol#1229) Co-authored-by: Tapan Chugh <tapanc@cs.washington.edu> * Improved supported for ProtectedResourceMetadata (modelcontextprotocol#1235) Co-authored-by: Paul Carleton <paulcarletonjr@gmail.com> * chore: Remove unused variable notification_options (modelcontextprotocol#1238) * Improve README around the Context object (modelcontextprotocol#1203) * fix: allow to pass `list[str]` to `token_endpoint_auth_signing_alg_values_supported` (modelcontextprotocol#1226) * Remove strict validation on `response_modes_supported` member of `OAuthMetadata` (modelcontextprotocol#1243) * Add pyright strict mode on the whole project (modelcontextprotocol#1254) * Consistent casing for default headers Accept and Content-Type (modelcontextprotocol#1263) * Update dependencies and fix type issues (modelcontextprotocol#1268) Co-authored-by: Marcelo Trylesinski <marcelotryle@gmail.com> * fix: prevent async generator cleanup errors in StreamableHTTP transport (modelcontextprotocol#1271) Co-authored-by: David Soria Parra <167242713+dsp-ant@users.noreply.github.com> * chore: uncomment .idea/ in .gitignore (modelcontextprotocol#1287) Co-authored-by: Claude <noreply@anthropic.com> * docs: clarify streamable_http_path configuration when mounting servers (modelcontextprotocol#1172) * feat: Add CORS configuration for browser-based MCP clients (modelcontextprotocol#1059) Co-authored-by: Marcelo Trylesinski <marcelotryle@gmail.com> Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> * Added Audio to FastMCP (modelcontextprotocol#1130) * fix: avoid uncessary retries in OAuth authenticated requests (modelcontextprotocol#1206) Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> * Add PATHEXT to default STDIO env vars in windows (modelcontextprotocol#1256) * fix: error too many values to unpack (expected 2) (modelcontextprotocol#1279) Signed-off-by: San Nguyen <vinhsannguyen91@gmail.com> Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * SDK Parity: Avoid Parsing Server Response for non-JsonRPCMessage Requests (modelcontextprotocol#1290) * types: Setting default value for method: Literal (modelcontextprotocol#1292) * changes structured temperature to not deadly (modelcontextprotocol#1328) * Update simple-resource example to use non-deprecated read_resource return type (modelcontextprotocol#1331) Co-authored-by: Claude <noreply@anthropic.com> * docs: Update README to include link to API docs for modelcontextprotocol#1329 (modelcontextprotocol#1330) * Allow ping requests before initialization (modelcontextprotocol#1312) * Python lint: Ruff rules for pylint and code complexity (modelcontextprotocol#525) * Fix context injection for resources and prompts (modelcontextprotocol#1336) * fix(fastmcp): propagate mimeType in resource template list (modelcontextprotocol#1186) Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * fix: allow elicitations accepted without content (modelcontextprotocol#1285) Co-authored-by: Olivier Schiavo <olivier.schiavo@wengo.com> * Use --frozen in pre-commit config (modelcontextprotocol#1375) * Return HTTP 403 for invalid Origin headers (modelcontextprotocol#1353) * Add test for ProtectedResourceMetadataParsing (modelcontextprotocol#1236) Co-authored-by: Paul Carleton <paulcarletonjr@gmail.com> Co-authored-by: Marcelo Trylesinski <marcelotryle@gmail.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * Fastmcp logging progress example (modelcontextprotocol#1270) Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * feat: add paginated list decorators for prompts, resources, and tools (modelcontextprotocol#1286) Co-authored-by: Claude <noreply@anthropic.com> * Remove "unconditionally" from conditional description (modelcontextprotocol#1289) * Use streamable-http consistently in examples (modelcontextprotocol#1389) * feat: Add SDK support for SEP-1034 default values in elicitation schemas (modelcontextprotocol#1337) Co-authored-by: Tapan Chugh <tapanc@cs.washington.edu> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * Implementation of SEP 973 - Additional metadata + icons support (modelcontextprotocol#1357) * Add error log for client stdio (modelcontextprotocol#924) Co-authored-by: Your Name <youremail@yourdomain.com> Co-authored-by: Marcelo Trylesinski <marcelotryle@gmail.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * Accept additional response_types values from OAuth servers (modelcontextprotocol#1323) * Issue 1379 patch - Fix MCP server OAuth not working with Visual Studio Code and others with extra grant_types (modelcontextprotocol#1380) * Add comprehensive Unicode tests for streamable HTTP transport (modelcontextprotocol#1381) * Update Icon.sizes to use string array format (modelcontextprotocol#1411) * Delete CODEOWNERS to eliminate notification overload (modelcontextprotocol#1413) * fix: fix the system message in simple-chatbot example (modelcontextprotocol#1394) * fix: improve misleading warning for progress callback exceptions (modelcontextprotocol#775) * fix: catch and rethrow SSEError during SSE connection establishment (modelcontextprotocol#975) Co-authored-by: zhangchuanhui <zhangchal@digitalchina.com> Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> * Add icons support for ResourceTemplate (modelcontextprotocol#1412) * Add documentation structure (modelcontextprotocol#1425) * Add documentation about testing (modelcontextprotocol#1426) * Improve OAuth protected resource metadata URL construction per RFC 9728 (modelcontextprotocol#1407) * feat: add ability to remove tools (modelcontextprotocol#1322) Co-authored-by: David Soria Parra <167242713+dsp-ant@users.noreply.github.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> Co-authored-by: Max Isbey <224885523+maxisbey@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com> * Update README to link to Python SDK documentation (modelcontextprotocol#1430) * fix: update CLAUDE.md to remove auto-addition of reviewers. (modelcontextprotocol#1431) * [client] Implement MCP OAuth scope selection and step-up authorization (modelcontextprotocol#1324) * Handles message type Exception in lowlevel/server.py _handle_message function. Mentioned as TODO on line 528. (modelcontextprotocol#786) Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * Fix workspace configuration error with structured_output_lowlevel.py (modelcontextprotocol#1471) Co-authored-by: lorenss-m <saeclmusic@gmail.com> * fix: Remove unnecessary constructor from ResourceServerSettings (modelcontextprotocol#1424) Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * feat: add resource annotations support to FastMCP (modelcontextprotocol#1468) * fix: send params as empty object for list methods without cursor (modelcontextprotocol#1453) * fix: Set the Server session initialization state immediately after respond… (modelcontextprotocol#1478) Co-authored-by: Max Isbey <224885523+maxisbey@users.noreply.github.com> * feat: add tool metadata in FastMCP.tool decorator (modelcontextprotocol#1463) Co-authored-by: Max Isbey <224885523+maxisbey@users.noreply.github.com> * Make client examples workspaces to reflect package code (modelcontextprotocol#1466) * Expose RequestParams._meta in ClientSession.call_tool (modelcontextprotocol#1231) Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * Allow CallToolResult to be returned directly to support _meta field for OpenAI Apps (modelcontextprotocol#1459) Co-authored-by: Max Isbey <224885523+maxisbey@users.noreply.github.com> * fix: uv CVE-2025-62518 astral-tokio-tar issue GHSA-j5gw-2vrg-8fgx (modelcontextprotocol#1505) * fix: use proper dependency resolution in CI (modelcontextprotocol#1507) * Upgrade GitHub Actions (modelcontextprotocol#1473) * test: use errno.ENOENT for command not found assertion (modelcontextprotocol#1498) * Replace deprecated dev-dependencies with dependency-groups (modelcontextprotocol#1488) Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * update uv to 0.9.5 (modelcontextprotocol#1510) * Relax Accept header requirement for JSON-only responses (modelcontextprotocol#1500) * fix: replace deprecated dev-dependencies in examples/clients (modelcontextprotocol#1518) * fix: Update spec links to new modelcontextprotocol.io location (modelcontextprotocol#1491) * fix: Replace fixed sleep with active server readiness check in SSE tests (modelcontextprotocol#1526) * fix: Replace arbitrary sleeps with active server readiness checks in tests (modelcontextprotocol#1527) Co-authored-by: Claude <noreply@anthropic.com> * Fix flaky timeout test in test_88_random_error (modelcontextprotocol#1525) * fix: Replace remaining manual server polling with wait_for_server helper (modelcontextprotocol#1529) * Implement RFC 7523 JWT flows (modelcontextprotocol#1247) Co-authored-by: Yann Jouanin <yann.jouanin@valueandco.com> * Fix pyright error and replace wildcard import with explicit imports (modelcontextprotocol#1532) * Fix auth client example URL handling for oauth provider (modelcontextprotocol#1549) * docs: use article "an" before "MCP" instead of "a" (modelcontextprotocol#1558) * Update Starlette to 0.49.1 in uv.lock (modelcontextprotocol#1559) * Fix typo in `ClientSessionGroup` doc string (modelcontextprotocol#1572) * Implement SEP-985: OAuth Protected Resource Metadata discovery fallback (modelcontextprotocol#1548) Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Paul Carleton <paulc@anthropic.com> * Add --frozen flag to uv run commands in Claude config (modelcontextprotocol#1583) * Add get_server_capabilities() to ClientSession (modelcontextprotocol#1588) * Add everything-server for comprehensive MCP conformance testing (modelcontextprotocol#1587) * Get baseline 100% clean coverage (modelcontextprotocol#1553) * Add end-of-file-fixer pre-commit hook (modelcontextprotocol#1610) * Add coverage baseline commit to git-blame-ignore (modelcontextprotocol#1613) * Add SEP-1034 conformance test support to everything-server (modelcontextprotocol#1604) Co-authored-by: Max Isbey <224885523+maxisbey@users.noreply.github.com> * refactor: extract OAuth helper functions and simplify provider state (modelcontextprotocol#1586) * Add client_id_metadata_document_supported to OAuthMetadata (modelcontextprotocol#1603) * Fix OAuth discovery fallback and URL ordering (modelcontextprotocol#1624) * Refactor `func_metadata()` implementation (modelcontextprotocol#1496) * Fix CI highest resolution test to actually test highest versions (modelcontextprotocol#1609) * feat: Pass through and expose additional parameters in `ClientSessionGroup.call_tool` and `.connect_to_server` (modelcontextprotocol#1576) * fix get_client_metadata_scopes on 401 (modelcontextprotocol#1631) Co-authored-by: Max Isbey <224885523+maxisbey@users.noreply.github.com> * chore: Lazy import `jsonschema` library (modelcontextprotocol#1596) Co-authored-by: Max Isbey <224885523+maxisbey@users.noreply.github.com> * docs: Update examples to use stateless HTTP with JSON responses (modelcontextprotocol#1499) * Add tests for JSON Schema 2020-12 field preservation (SEP-1613) (modelcontextprotocol#1649) * Add client_secret_basic authentication support (modelcontextprotocol#1334) Co-authored-by: Paul Carleton <paulc@anthropic.com> * Implement SEP-1577 - Sampling With Tools (modelcontextprotocol#1594) Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> Co-authored-by: Claude <noreply@anthropic.com> * SEP-1330: Elicitation Enum Schema Improvements and Standards Compliance (modelcontextprotocol#1246) Co-authored-by: Tapan Chugh <tapanc@cs.washington.edu> Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * [auth][conformance] add conformance auth client (modelcontextprotocol#1640) Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * Implement SEP-986: Tool name validation (modelcontextprotocol#1655) * fix: url for spec (modelcontextprotocol#1659) * feat: implement SEP-991 URL-based client ID (CIMD) support (modelcontextprotocol#1652) Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * Update doc string on custom_route (modelcontextprotocol#1660) * Implement SEP-1036: URL mode elicitation for secure out-of-band interactions (modelcontextprotocol#1580) Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> * Skip empty SSE data to avoid parsing errors (modelcontextprotocol#1670) * SEP-1686: Tasks (modelcontextprotocol#1645) * Add on_session_created callback option (modelcontextprotocol#1710) * Add SSE polling support (SEP-1699) (modelcontextprotocol#1654) * Support client_credentials flow with JWT and Basic auth (modelcontextprotocol#1663) Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> * feat: backwards-compatible create_message overloads for SEP-1577 (modelcontextprotocol#1713) * Merge commit from fork * Auto-enable DNS rebinding protection for localhost servers When a FastMCP server is created with host="127.0.0.1" or "localhost" and no explicit transport_security is provided, automatically enable DNS rebinding protection. Both 127.0.0.1 and localhost are allowed as valid hosts/origins since clients may use either to connect. * Add tests for auto DNS rebinding protection on localhost Tests verify that: - Protection auto-enables for host=127.0.0.1 - Protection auto-enables for host=localhost - Both 127.0.0.1 and localhost are in allowed hosts/origins - Protection does NOT auto-enable for other hosts (e.g., 0.0.0.0) - Explicit transport_security settings are not overridden * Add IPv6 localhost (::1) support for DNS rebinding protection Extend auto-enable DNS rebinding protection to also cover IPv6 localhost. When host="::1", protection is now auto-enabled with appropriate allowed hosts ([::1]:*) and origins (http://[::1]:*). * Fix import ordering in test file * chore: update LATEST_PROTOCOL_VERSION to 2025-11-25 (modelcontextprotocol#1715) * fix: add lifespan context manager to StreamableHTTP mounting examples (modelcontextprotocol#1669) Co-authored-by: TheMailmans <tyler@example.com> Co-authored-by: Marcelo Trylesinski <marcelotryle@gmail.com> * fix: handle ClosedResourceError in StreamableHTTP message router (modelcontextprotocol#1384) Co-authored-by: Max Isbey <224885523+maxisbey@users.noreply.github.com> Co-authored-by: Marcelo Trylesinski <marcelotryle@gmail.com> * fix: skip priming events and close_sse_stream for old protocol versions (modelcontextprotocol#1719) * refactor(auth): remove unused _register_client method (modelcontextprotocol#1748) * [MCP-266] Add tests for Gumloop server extensions * Fix uv workspace config for gumloop-mcp package name * Sync with upstream MCP SDK and fix merge conflicts * Fix tool cache timing and missing properties check in server.py * Fix coverage and add proper type annotations for Gumloop extensions * Version up * Skip README code example tests (Gumloop README has no code snippets) * Support gumloop and mcp outptuschema * Add publish tools to dev dependencies and update README for uv --------- Signed-off-by: San Nguyen <vinhsannguyen91@gmail.com> Co-authored-by: Felix Weinberger <3823880+felixweinberger@users.noreply.github.com> Co-authored-by: yurikunash <143175350+yurikunash@users.noreply.github.com> Co-authored-by: Pamela Fox <pamela.fox@gmail.com> Co-authored-by: Inna Harper <inna.hrpr@gmail.com> Co-authored-by: Marcelo Trylesinski <marcelotryle@gmail.com> Co-authored-by: Ian Davenport <49379192+davenpi@users.noreply.github.com> Co-authored-by: Dagang Wei <functicons@gmail.com> Co-authored-by: Felix Weinberger <fweinberger@anthropic.com> Co-authored-by: Stanley Law <stanleylkal@gmail.com> Co-authored-by: Luca Chang <131398524+LucaButBoring@users.noreply.github.com> Co-authored-by: leweng <leweng@nvidia.com> Co-authored-by: Clare Liguori <liguori@amazon.com> Co-authored-by: lukacf <luka@peltarion.com> Co-authored-by: ihrpr <inna@anthropic.com> Co-authored-by: Tapan Chugh <chugh.tapan@gmail.com> Co-authored-by: Tapan Chugh <tapanc@cs.washington.edu> Co-authored-by: Yann Jouanin <4557670+yannj-fr@users.noreply.github.com> Co-authored-by: Paul Carleton <paulcarletonjr@gmail.com> Co-authored-by: Sreenath Somarajapuram <somarajapuram@gmail.com> Co-authored-by: Omer Korner <omerkorner@gmail.com> Co-authored-by: joesavage-silabs <159480754+joesavage-silabs@users.noreply.github.com> Co-authored-by: Gregory L <gregory.linford@mistral.ai> Co-authored-by: David Soria Parra <167242713+dsp-ant@users.noreply.github.com> Co-authored-by: Moustapha Ebnou <155577789+mous222@users.noreply.github.com> Co-authored-by: Max Isbey <224885523+maxisbey@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Jerome <jerome@anthropic.com> Co-authored-by: xavier <84836280+dragonier23@users.noreply.github.com> Co-authored-by: keurcien <keurcien.luu@gmail.com> Co-authored-by: Tim Esler <tim.esler@gmail.com> Co-authored-by: San Nguyen <22189661+sandangel@users.noreply.github.com> Co-authored-by: Justin Wang <89049861+justin-yi-wang@users.noreply.github.com> Co-authored-by: jess <jessachandler@gmail.com> Co-authored-by: Peter Alexander <pja@anthropic.com> Co-authored-by: Reid Geyer <12072650+reidg44@users.noreply.github.com> Co-authored-by: Eleftheria Stein-Kousathana <eleftheria.kousathana@gmail.com> Co-authored-by: Christian Clauss <cclauss@me.com> Co-authored-by: pchoudhury22 <pchoudhury22@apple.com> Co-authored-by: owengo <owengo@users.noreply.github.com> Co-authored-by: Olivier Schiavo <olivier.schiavo@wengo.com> Co-authored-by: Steve Billings <billings.steve@gmail.com> Co-authored-by: Mike Salvatore <mike.s.salvatore@gmail.com> Co-authored-by: pengwa <pengwa@microsoft.com> Co-authored-by: Your Name <youremail@yourdomain.com> Co-authored-by: Jon Shea <jonshea@jonshea.com> Co-authored-by: automaton82 <terrence.sheflin@gmail.com> Co-authored-by: Yukuan Jia <jiayukuan@huawei.com> Co-authored-by: Lorenzo <lorenzo_cesconeto@hotmail.com> Co-authored-by: ZhangChuanhui <57099533+zhangch-ss@users.noreply.github.com> Co-authored-by: zhangchuanhui <zhangchal@digitalchina.com> Co-authored-by: Marcus Shu <46469249+shulkx@users.noreply.github.com> Co-authored-by: Brandon Wu <49291449+brandonspark@users.noreply.github.com> Co-authored-by: Dogacan Colak <dogacancolak@gmail.com> Co-authored-by: AishwaryaKalloli <30429206+AishwaryaKalloli@users.noreply.github.com> Co-authored-by: lorenss-m <saeclmusic@gmail.com> Co-authored-by: Rocky Haotian Du <2712479005@qq.com> Co-authored-by: Fenn Bailey <fennb@users.noreply.github.com> Co-authored-by: daamitt <147169947+daamitt@users.noreply.github.com> Co-authored-by: Mat Leonard <137834585+mat-octave@users.noreply.github.com> Co-authored-by: Samuel Felipe Chenatti <samuel.chenatti@gmail.com> Co-authored-by: Brandon Shar <6599653+BrandonShar@users.noreply.github.com> Co-authored-by: mingo007 <maoqiming1@huawei.com> Co-authored-by: adam jones <domdomegg+git@gmail.com> Co-authored-by: Yann Jouanin <yann.jouanin@valueandco.com> Co-authored-by: Koichi ITO <koic.ito@gmail.com> Co-authored-by: Cole Murray <colemurray.cs@gmail.com> Co-authored-by: inaku <63503085+inaku-Gyan@users.noreply.github.com> Co-authored-by: Chris Coutinho <12901868+cbcoutinho@users.noreply.github.com> Co-authored-by: Paul Carleton <paulc@anthropic.com> Co-authored-by: Camila Rondinini <camila@anthropic.com> Co-authored-by: Victorien <65306057+Viicos@users.noreply.github.com> Co-authored-by: Andrii Blyzniuk <bliznyuk.andrey@gmail.com> Co-authored-by: Liang Wu <18244712+wuliang229@users.noreply.github.com> Co-authored-by: adam jones <adamj+git@anthropic.com> Co-authored-by: Olivier Chafik <ochafik@anthropic.com> Co-authored-by: Tyler Mailman <themailmaninbox@gmail.com> Co-authored-by: TheMailmans <tyler@example.com> Co-authored-by: Edison <Edison.A.N@hotmail.com> Co-authored-by: dvlpjrs <dvlp.jrs@gmail.com>
maxisbey
added a commit
that referenced
this pull request
Mar 18, 2026
When the transport closes (stdin EOF, client disconnect) while a request handler is still running, server.run()'s task group joins on the handler instead of cancelling it. The handler eventually finishes, tries to send its response through a write stream that _receive_loop already closed, and server.run() crashes with ClosedResourceError wrapped in a triple-nested ExceptionGroup. The fix cancels the task group when the incoming_messages loop ends. Handlers receive CancelledError and can clean up in finally blocks. The existing CancelledError catch in _handle_request (added for CancelledNotification handling in #1153) now distinguishes the two cancellation sources: responder.cancel() already sent an error response and we skip the duplicate; transport-close cancellation is re-raised so the task group swallows it. Github-Issue: #526
9 tasks
QuentinAmbard
pushed a commit
to databricks-solutions/ai-dev-kit
that referenced
this pull request
Apr 2, 2026
When a client cancels a long-running MCP request, there's a race condition between the cancellation and normal response paths: 1. Client cancels request → RequestResponder.cancel() sends error response and sets _completed = True 2. Middleware catches CancelledError and returns a ToolResult 3. MCP SDK tries to call message.respond(response) 4. Crash: assert not self._completed fails Fix: Re-raise CancelledError instead of returning a result, allowing the MCP SDK's cancellation handler to properly manage the response lifecycle. See: modelcontextprotocol/python-sdk#1153 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
4 tasks
calreynolds
pushed a commit
to databricks-solutions/ai-dev-kit
that referenced
this pull request
Apr 2, 2026
#411) * Fix MCP server crash on request cancellation When a client cancels a long-running MCP request, there's a race condition between the cancellation and normal response paths: 1. Client cancels request → RequestResponder.cancel() sends error response and sets _completed = True 2. Middleware catches CancelledError and returns a ToolResult 3. MCP SDK tries to call message.respond(response) 4. Crash: assert not self._completed fails Fix: Re-raise CancelledError instead of returning a result, allowing the MCP SDK's cancellation handler to properly manage the response lifecycle. See: modelcontextprotocol/python-sdk#1153 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add structured_content to error responses for MCP SDK validation When tools have an outputSchema (auto-generated from return type like Dict[str, Any]), MCP SDK requires structured_content in all responses. The middleware was returning ToolResult without structured_content for error cases (timeout, exceptions), causing validation errors: "Output validation error: outputSchema defined but no structured output returned" Fix: Include structured_content with the same error data in all error responses. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix structured_content not populated for tools with return type annotations FastMCP auto-generates outputSchema from return type annotations (e.g., -> Dict[str, Any]) but doesn't populate structured_content in ToolResult. MCP SDK validation then fails: "outputSchema defined but no structured output" Fix: Intercept successful results and populate structured_content from JSON text content when missing. Only modifies results when: 1. structured_content is missing 2. There's exactly one TextContent item 3. The text is valid JSON that parses to a dict 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * fix(mcp): apply async wrapper on all platforms to prevent cancellation crashes The asyncio.to_thread() wrapper was only applied on Windows, but it's needed on ALL platforms to enable proper cancellation handling. Without this fix, when a sync tool runs longer than the client timeout: 1. Client sends cancellation 2. Sync tool blocks event loop, can't receive CancelledError 3. Tool eventually returns, but MCP SDK already responded to cancel 4. AssertionError: "Request already responded to" → server crashes This was discovered when uploading 7,375 files triggered a timeout, crashing the MCP server on macOS. Extends the fix from PR #411 which added CancelledError handling in middleware - that fix only works when cancellation can propagate, which requires async execution via to_thread(). 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix: don't set structured_content on error responses Setting structured_content causes MCP SDK to validate it against the tool's outputSchema. For error responses, the error dict {"error": True, ...} doesn't match the expected return type (e.g., Union[str, List[Dict]]), causing "Output validation error: 'result' is a required property". Fix: Only set structured_content for successful responses, not errors. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Quentin Ambard <quentin.ambard@databricks.com> Co-authored-by: Claude <noreply@anthropic.com>
calreynolds
added a commit
to databricks-solutions/ai-dev-kit
that referenced
this pull request
Apr 8, 2026
…#374) * Enhance AI/BI dashboard skill with comprehensive widget specs Added missing documentation from production dashboard generation: 1-widget-specifications.md: - Combo charts (bar + line on same widget) with version 1 - Counter number formatting (currency, percent, plain number) - Widget name max length (60 characters) - Color scale restrictions (no scheme/colorRamp/mappings) - Quantitative color encoding for gradient effects - Bar chart group vs stacked decision criteria with examples 2-filters.md: - Date range picker complete example - Multi-dataset filter binding (one query per dataset) - Global filter performance note (auto WHERE clause) SKILL.md: - ORDER BY guidance for time series and rankings 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add TOP-N + Other bucketing guidance for high-cardinality dimensions When a dimension has too many values (50+ stores, products, etc.), charts become unreadable. Added guidance to: - Check cardinality via get_table_details before charting - Use TOP-N + "Other" SQL pattern to bucket low-value items - Aggregate to higher abstraction level as alternative - Use table widgets for high-cardinality data 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Remove async deploy_dashboard function for consistency The codebase doesn't use async anywhere else, so remove the unused async version of deploy_dashboard and keep only the synchronous one. - Remove asyncio import - Remove async deploy_dashboard function (was using asyncio.to_thread) - Rename deploy_dashboard_sync to deploy_dashboard - Update exports in __init__.py 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add genie_space_id parameter to dashboard creation Allow linking a Genie space to a dashboard by passing genie_space_id. This enables the "Ask Genie" button on the dashboard UI. The Genie space config is injected into the serialized_dashboard JSON under uiSettings.genieSpace with isEnabled=true and enablementMode=ENABLED. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add catalog and schema parameters to dashboard creation Allow setting default catalog and schema for dashboard datasets via the dataset_catalog and dataset_schema API parameters. These defaults apply to unqualified table names in SQL queries. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add comprehensive date range filtering documentation - Document field-based filtering (automatic IN_RANGE on date fields) - Document parameter-based filtering (:date_range.min/max in SQL) - Show how to combine both approaches in one filter - Add guidance on when NOT to apply date filtering (MRR, all-time totals) - Update SKILL.md tools table with new genie_space_id, catalog, schema params 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Restructure AI/BI dashboard skill with improved organization - Split widget specs into basic (1-widget-specifications.md) and advanced (2-advanced-widget-specifications.md) files - Add area chart, scatter plot, combo chart, and choropleth map documentation - Rename files for consistent numbering (3-filters, 4-examples, 5-troubleshooting) - Remove duplicate information across files (versions, naming rules, etc.) - Add widget display formatting guidance (currency, percentage, displayName) - Simplify SKILL.md quality checklist with link to version table - Shorten verbose examples while preserving all critical information - Clarify query naming convention for charts vs filters 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add back critical behavioral instructions for text widgets and filters * Restore important behavioral instructions removed during restructure - Counter: full Pattern 2 example with CRITICAL field name matching note - Table: disaggregated:true guidance and bold emphasis - Line/Bar: x,y,color encodings and disaggregated guidance - Pie: 3-8 category limit for readability * Restore detailed guidance that was removed during restructure - 5-troubleshooting.md: Restore full troubleshooting content with version guidance, filter debugging, and detailed error explanations - SKILL.md: Restore full 10-item quality checklist - SKILL.md: Restore standard dashboard structure example - SKILL.md: Restore cardinality guidance table (with softer 'suggested' language) * Optimize MCP tool docstrings for token efficiency - Reduce docstring verbosity across all 18 tool files (~89% reduction) - Keep all functional information while being concise - Add skill references to complex tools (dashboards, vector search, genie, jobs, pipelines, lakebase, unity catalog, serving, apps, agent bricks) - Maintain human readability with bullet points and structure - Preserve critical warnings (ASK USER FIRST, CONFIRM WITH USER) - Keep return format hints for AI parsing Net reduction: 1,843 lines across 18 files 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add parameter context for ambiguous docstring params - agent_bricks.py: Add context for description, instructions, volume_path, examples - genie.py: Add context for table_identifiers, description, sample_questions, serialized_space - jobs.py: Add context for tasks, job_clusters, environments, schedule, git_source - lakebase.py: Add context for source_branch, ttl_seconds, is_protected, autoscaling params, and sync source/target table names - pipelines.py: Add context for root_path, workspace_file_paths, extra_settings, full_refresh 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Consolidate MCP tools from 77 to 44 (43% reduction) Tool consolidations: - pipelines.py: 10→2 (manage_pipeline, manage_pipeline_run) - volume_files.py: 6→1 (manage_volume_files) - aibi_dashboards.py: 4→1 (manage_dashboard) - vector_search.py: 8→4 (manage_vs_endpoint, manage_vs_index, query_vs_index, manage_vs_data) - genie.py: 5→2 (manage_genie, ask_genie) - serving.py: 3→1 (manage_serving_endpoint) - apps.py: 3→1 (manage_app) - file.py: 2→1 (manage_workspace_files) - sql.py: 6→5 (manage_warehouse replaces list/get_best) - lakebase.py: 8→4 (manage_lakebase_database, manage_lakebase_branch, manage_lakebase_sync, generate_lakebase_credential) Key patterns: - All consolidated tools use an action parameter - Each action has required params documented in docstring - Error messages specify which params are required - Hot paths (query_vs_index, ask_genie) kept separate for clarity - All skills updated with action tables and examples 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add integration test infrastructure and fix tool bugs Test infrastructure: - Add comprehensive integration tests for all MCP tools - Add test runner script with parallel execution support - Add fixtures for workspace, catalog, and resource cleanup - Add test resources (PDFs, SQL files, app configs) Bug fixes in databricks-tools-core: - Fix workspace file upload for directories - Fix job notebook path handling - Fix vector search index operations - Fix apps API responses - Fix dashboard widget handling - Fix agent bricks manager listing Bug fixes in MCP server tools: - Add quota skip handling for apps test - Fix genie space operations - Fix lakebase database operations - Fix compute cluster lifecycle handling - Fix dashboard operations 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Change manage_dashboard to use file path instead of inline JSON - Replace serialized_dashboard param with dashboard_file_path - Tool reads JSON from local file for easier iterative development - Update SKILL.md with new workflow documentation 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Update dashboard tests for file-based approach - Change simple_dashboard_json fixture to simple_dashboard_file - Update all manage_dashboard calls to use dashboard_file_path - Add tempfile imports and tmp_path usage for update test 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix deploy_app to correctly handle SDK Wait[AppDeployment] return type The Databricks SDK's w.apps.deploy() returns a Wait[AppDeployment] object, not an AppDeployment directly. The previous code passed the Wait object to _deployment_to_dict(), which caused getattr() to return None for all attributes since the Wait object doesn't have them. This fix uses wait_obj.response to get the actual AppDeployment object before converting it to a dictionary. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix deploy_app to correctly handle SDK Wait[AppDeployment] return type The Databricks SDK's w.apps.deploy() returns a Wait[AppDeployment] object, not an AppDeployment directly. The previous code passed the Wait object to _deployment_to_dict(), which caused getattr() to return None for all attributes since the Wait object doesn't have them. This fix uses wait_obj.response to get the actual AppDeployment object before converting it to a dictionary. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Clarify MCP tool usage in Genie skill documentation - Add tools summary table at top of MCP Tools section - Change code blocks from python syntax to plain text - Add "# MCP Tool: <name>" comments to clarify these are tool calls, not Python code - Move Supporting Tools table to main tools table 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix typo in aibi_dashboards.py docstring Remove garbage characters from widget documentation. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix genie tools to use SDK methods instead of manager - Use w.genie.trash_space() in _delete_genie_resource - Add _find_space_by_name() using SDK's list_spaces with pagination - Use w.genie.update_space() and w.genie.create_space() for space management - Use w.genie.get_space() with include_serialized_space in _get_genie_space - Fix validation to allow space_id for updates without display_name 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Improve integration test reliability and timeout handling - Add per-suite timeout in run_tests.py (10 min default, configurable) - Improve apps test with better cleanup and assertions - Add skip logic for quota-exceeded scenarios 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Improve Unity Catalog tool docstrings with comprehensive parameter documentation Add detailed parameter documentation to all 9 Unity Catalog MCP tools: - manage_uc_objects: Document parameters by object_type (catalog/schema/volume/function) - manage_uc_grants: Add privilege lists per securable type - manage_uc_storage: Detail credential and external_location parameters - manage_uc_connections: Document connection_type options and create_foreign_catalog - manage_uc_tags: Detail set_tags/unset_tags/query parameters - manage_uc_security_policies: Document row filter and column mask parameters - manage_uc_monitors: Detail monitor creation and refresh parameters - manage_uc_sharing: Document share/recipient/provider resource types - manage_metric_views: Detail dimension/measure format and query parameters 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add CRITICAL validation steps to dashboard tool docstring Add clear instructions requiring users to: 0. Review the databricks-aibi-dashboards skill for widget JSON structure 1. Call get_table_stats_and_schema() for table schemas 2. Call execute_sql() to test EVERY query before use This prevents widgets from showing errors due to untested queries. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add design best practices section and use relative file paths - Add Design Best Practices section for default dashboard behaviors - Change /tmp paths to ./ for less opinionated examples - Update parent_path example to use {user_email} placeholder 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Improve AI/BI dashboard skill documentation with comprehensive examples - Replace basic NYC taxi examples with complete Sales Analytics dashboard - Add critical widget version requirements table to SKILL.md - Add data validation guidance to verify dashboards tell intended story - Document key patterns: page types, KPI formatting, filter binding, layout grid 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add skill reading requirement to dashboard MCP tool docstring Require agent to read 4-examples.md before creating dashboards, and if unfamiliar, read full skill documentation first. Valid JSON is critical. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix MCP server crash on request cancellation When a client cancels a long-running MCP request, there's a race condition between the cancellation and normal response paths: 1. Client cancels request → RequestResponder.cancel() sends error response and sets _completed = True 2. Middleware catches CancelledError and returns a ToolResult 3. MCP SDK tries to call message.respond(response) 4. Crash: assert not self._completed fails Fix: Re-raise CancelledError instead of returning a result, allowing the MCP SDK's cancellation handler to properly manage the response lifecycle. See: modelcontextprotocol/python-sdk#1153 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add structured_content to error responses for MCP SDK validation When tools have an outputSchema (auto-generated from return type like Dict[str, Any]), MCP SDK requires structured_content in all responses. The middleware was returning ToolResult without structured_content for error cases (timeout, exceptions), causing validation errors: "Output validation error: outputSchema defined but no structured output returned" Fix: Include structured_content with the same error data in all error responses. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Migrate KA operations to Python SDK and fix name lookup issues - Migrate ka_create, ka_get, ka_sync_sources to use Python SDK - Keep ka_update using raw API 2.1 due to SDK FieldMask bug (converts snake_case to camelCase but API expects snake_case) - Fix find_by_name to sanitize names (spaces→underscores) before lookup - Fix ka_create_or_update to lookup by name when no tile_id provided, preventing ALREADY_EXISTS errors on repeated calls - Update MCP tool layer to use new flat response format - Map SDK state values (ACTIVE, CREATING, FAILED) to endpoint_status - Add integration test for updating existing KA via create_or_update 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix knowledge source description requirement and test ordering - Provide default description for knowledge sources when not specified (API requires non-empty knowledge_source.description) - Move KA update test to after endpoint is ONLINE (update requires ACTIVE state) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix structured_content not populated for tools with return type annotations FastMCP auto-generates outputSchema from return type annotations (e.g., -> Dict[str, Any]) but doesn't populate structured_content in ToolResult. MCP SDK validation then fails: "outputSchema defined but no structured output" Fix: Intercept successful results and populate structured_content from JSON text content when missing. Only modifies results when: 1. structured_content is missing 2. There's exactly one TextContent item 3. The text is valid JSON that parses to a dict 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * fix(mcp): apply async wrapper on all platforms to prevent cancellation crashes The asyncio.to_thread() wrapper was only applied on Windows, but it's needed on ALL platforms to enable proper cancellation handling. Without this fix, when a sync tool runs longer than the client timeout: 1. Client sends cancellation 2. Sync tool blocks event loop, can't receive CancelledError 3. Tool eventually returns, but MCP SDK already responded to cancel 4. AssertionError: "Request already responded to" → server crashes This was discovered when uploading 7,375 files triggered a timeout, crashing the MCP server on macOS. Extends the fix from PR #411 which added CancelledError handling in middleware - that fix only works when cancellation can propagate, which requires async execution via to_thread(). 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix: don't set structured_content on error responses Setting structured_content causes MCP SDK to validate it against the tool's outputSchema. For error responses, the error dict {"error": True, ...} doesn't match the expected return type (e.g., Union[str, List[Dict]]), causing "Output validation error: 'result' is a required property". Fix: Only set structured_content for successful responses, not errors. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Improve dashboard skill structure based on error analysis - Add JSON skeleton section to SKILL.md showing required structure - Add Genie note clarifying it's not a widget (use genie_space_id param) - Move Key Patterns to top of 4-examples.md for discoverability - Clarify example is reference only - adapt to user's actual requirements - Add structural errors table to 5-troubleshooting.md Root cause fixes: - queryLines must be array, not "query": "string" - Widgets must be inline in layout[].widget, not separate array - pageType required on every page 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Quentin Ambard <quentin.ambard@databricks.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: calreynolds <calrey98@gmail.com>
calreynolds
added a commit
to databricks-solutions/ai-dev-kit
that referenced
this pull request
Apr 8, 2026
* Enhance AI/BI dashboard skill with comprehensive widget specs Added missing documentation from production dashboard generation: 1-widget-specifications.md: - Combo charts (bar + line on same widget) with version 1 - Counter number formatting (currency, percent, plain number) - Widget name max length (60 characters) - Color scale restrictions (no scheme/colorRamp/mappings) - Quantitative color encoding for gradient effects - Bar chart group vs stacked decision criteria with examples 2-filters.md: - Date range picker complete example - Multi-dataset filter binding (one query per dataset) - Global filter performance note (auto WHERE clause) SKILL.md: - ORDER BY guidance for time series and rankings 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add TOP-N + Other bucketing guidance for high-cardinality dimensions When a dimension has too many values (50+ stores, products, etc.), charts become unreadable. Added guidance to: - Check cardinality via get_table_details before charting - Use TOP-N + "Other" SQL pattern to bucket low-value items - Aggregate to higher abstraction level as alternative - Use table widgets for high-cardinality data 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Remove async deploy_dashboard function for consistency The codebase doesn't use async anywhere else, so remove the unused async version of deploy_dashboard and keep only the synchronous one. - Remove asyncio import - Remove async deploy_dashboard function (was using asyncio.to_thread) - Rename deploy_dashboard_sync to deploy_dashboard - Update exports in __init__.py 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add genie_space_id parameter to dashboard creation Allow linking a Genie space to a dashboard by passing genie_space_id. This enables the "Ask Genie" button on the dashboard UI. The Genie space config is injected into the serialized_dashboard JSON under uiSettings.genieSpace with isEnabled=true and enablementMode=ENABLED. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add catalog and schema parameters to dashboard creation Allow setting default catalog and schema for dashboard datasets via the dataset_catalog and dataset_schema API parameters. These defaults apply to unqualified table names in SQL queries. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add comprehensive date range filtering documentation - Document field-based filtering (automatic IN_RANGE on date fields) - Document parameter-based filtering (:date_range.min/max in SQL) - Show how to combine both approaches in one filter - Add guidance on when NOT to apply date filtering (MRR, all-time totals) - Update SKILL.md tools table with new genie_space_id, catalog, schema params 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Restructure AI/BI dashboard skill with improved organization - Split widget specs into basic (1-widget-specifications.md) and advanced (2-advanced-widget-specifications.md) files - Add area chart, scatter plot, combo chart, and choropleth map documentation - Rename files for consistent numbering (3-filters, 4-examples, 5-troubleshooting) - Remove duplicate information across files (versions, naming rules, etc.) - Add widget display formatting guidance (currency, percentage, displayName) - Simplify SKILL.md quality checklist with link to version table - Shorten verbose examples while preserving all critical information - Clarify query naming convention for charts vs filters 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add back critical behavioral instructions for text widgets and filters * Restore important behavioral instructions removed during restructure - Counter: full Pattern 2 example with CRITICAL field name matching note - Table: disaggregated:true guidance and bold emphasis - Line/Bar: x,y,color encodings and disaggregated guidance - Pie: 3-8 category limit for readability * Restore detailed guidance that was removed during restructure - 5-troubleshooting.md: Restore full troubleshooting content with version guidance, filter debugging, and detailed error explanations - SKILL.md: Restore full 10-item quality checklist - SKILL.md: Restore standard dashboard structure example - SKILL.md: Restore cardinality guidance table (with softer 'suggested' language) * Optimize MCP tool docstrings for token efficiency - Reduce docstring verbosity across all 18 tool files (~89% reduction) - Keep all functional information while being concise - Add skill references to complex tools (dashboards, vector search, genie, jobs, pipelines, lakebase, unity catalog, serving, apps, agent bricks) - Maintain human readability with bullet points and structure - Preserve critical warnings (ASK USER FIRST, CONFIRM WITH USER) - Keep return format hints for AI parsing Net reduction: 1,843 lines across 18 files 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add parameter context for ambiguous docstring params - agent_bricks.py: Add context for description, instructions, volume_path, examples - genie.py: Add context for table_identifiers, description, sample_questions, serialized_space - jobs.py: Add context for tasks, job_clusters, environments, schedule, git_source - lakebase.py: Add context for source_branch, ttl_seconds, is_protected, autoscaling params, and sync source/target table names - pipelines.py: Add context for root_path, workspace_file_paths, extra_settings, full_refresh 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Consolidate MCP tools from 77 to 44 (43% reduction) Tool consolidations: - pipelines.py: 10→2 (manage_pipeline, manage_pipeline_run) - volume_files.py: 6→1 (manage_volume_files) - aibi_dashboards.py: 4→1 (manage_dashboard) - vector_search.py: 8→4 (manage_vs_endpoint, manage_vs_index, query_vs_index, manage_vs_data) - genie.py: 5→2 (manage_genie, ask_genie) - serving.py: 3→1 (manage_serving_endpoint) - apps.py: 3→1 (manage_app) - file.py: 2→1 (manage_workspace_files) - sql.py: 6→5 (manage_warehouse replaces list/get_best) - lakebase.py: 8→4 (manage_lakebase_database, manage_lakebase_branch, manage_lakebase_sync, generate_lakebase_credential) Key patterns: - All consolidated tools use an action parameter - Each action has required params documented in docstring - Error messages specify which params are required - Hot paths (query_vs_index, ask_genie) kept separate for clarity - All skills updated with action tables and examples 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add integration test infrastructure and fix tool bugs Test infrastructure: - Add comprehensive integration tests for all MCP tools - Add test runner script with parallel execution support - Add fixtures for workspace, catalog, and resource cleanup - Add test resources (PDFs, SQL files, app configs) Bug fixes in databricks-tools-core: - Fix workspace file upload for directories - Fix job notebook path handling - Fix vector search index operations - Fix apps API responses - Fix dashboard widget handling - Fix agent bricks manager listing Bug fixes in MCP server tools: - Add quota skip handling for apps test - Fix genie space operations - Fix lakebase database operations - Fix compute cluster lifecycle handling - Fix dashboard operations 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Change manage_dashboard to use file path instead of inline JSON - Replace serialized_dashboard param with dashboard_file_path - Tool reads JSON from local file for easier iterative development - Update SKILL.md with new workflow documentation 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Update dashboard tests for file-based approach - Change simple_dashboard_json fixture to simple_dashboard_file - Update all manage_dashboard calls to use dashboard_file_path - Add tempfile imports and tmp_path usage for update test 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix deploy_app to correctly handle SDK Wait[AppDeployment] return type The Databricks SDK's w.apps.deploy() returns a Wait[AppDeployment] object, not an AppDeployment directly. The previous code passed the Wait object to _deployment_to_dict(), which caused getattr() to return None for all attributes since the Wait object doesn't have them. This fix uses wait_obj.response to get the actual AppDeployment object before converting it to a dictionary. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix deploy_app to correctly handle SDK Wait[AppDeployment] return type The Databricks SDK's w.apps.deploy() returns a Wait[AppDeployment] object, not an AppDeployment directly. The previous code passed the Wait object to _deployment_to_dict(), which caused getattr() to return None for all attributes since the Wait object doesn't have them. This fix uses wait_obj.response to get the actual AppDeployment object before converting it to a dictionary. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Clarify MCP tool usage in Genie skill documentation - Add tools summary table at top of MCP Tools section - Change code blocks from python syntax to plain text - Add "# MCP Tool: <name>" comments to clarify these are tool calls, not Python code - Move Supporting Tools table to main tools table 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix typo in aibi_dashboards.py docstring Remove garbage characters from widget documentation. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix genie tools to use SDK methods instead of manager - Use w.genie.trash_space() in _delete_genie_resource - Add _find_space_by_name() using SDK's list_spaces with pagination - Use w.genie.update_space() and w.genie.create_space() for space management - Use w.genie.get_space() with include_serialized_space in _get_genie_space - Fix validation to allow space_id for updates without display_name 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Improve integration test reliability and timeout handling - Add per-suite timeout in run_tests.py (10 min default, configurable) - Improve apps test with better cleanup and assertions - Add skip logic for quota-exceeded scenarios 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Improve Unity Catalog tool docstrings with comprehensive parameter documentation Add detailed parameter documentation to all 9 Unity Catalog MCP tools: - manage_uc_objects: Document parameters by object_type (catalog/schema/volume/function) - manage_uc_grants: Add privilege lists per securable type - manage_uc_storage: Detail credential and external_location parameters - manage_uc_connections: Document connection_type options and create_foreign_catalog - manage_uc_tags: Detail set_tags/unset_tags/query parameters - manage_uc_security_policies: Document row filter and column mask parameters - manage_uc_monitors: Detail monitor creation and refresh parameters - manage_uc_sharing: Document share/recipient/provider resource types - manage_metric_views: Detail dimension/measure format and query parameters 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add CRITICAL validation steps to dashboard tool docstring Add clear instructions requiring users to: 0. Review the databricks-aibi-dashboards skill for widget JSON structure 1. Call get_table_stats_and_schema() for table schemas 2. Call execute_sql() to test EVERY query before use This prevents widgets from showing errors due to untested queries. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add design best practices section and use relative file paths - Add Design Best Practices section for default dashboard behaviors - Change /tmp paths to ./ for less opinionated examples - Update parent_path example to use {user_email} placeholder 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Improve code execution and synthetic data gen skills Execution compute skill: - Split into 3 clear execution modes: Databricks Connect (default), Serverless Job, Interactive Cluster - Add decision matrix for choosing execution mode - Add job_extra_params for custom dependencies in serverless jobs - Create dedicated reference files for each mode Synthetic data gen skill: - Emphasize business story: problem → impact → analysis → solution - Add guidance to propose compelling stories by default - Consolidate reference files (6 → 2) - Add critical rules for data coherence and Databricks value - Clarify when to read reference files 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add performance rules for synthetic data generation - Add Critical Rule #12: No Python loops or .collect() - Add Performance Rules section with anti-pattern table - Emphasize Spark parallelism over driver-side iteration 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Clarify parquet path is folder with table name 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Improve AI/BI dashboard skill documentation with comprehensive examples - Replace basic NYC taxi examples with complete Sales Analytics dashboard - Add critical widget version requirements table to SKILL.md - Add data validation guidance to verify dashboards tell intended story - Document key patterns: page types, KPI formatting, filter binding, layout grid 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add skill reading requirement to dashboard MCP tool docstring Require agent to read 4-examples.md before creating dashboards, and if unfamiliar, read full skill documentation first. Valid JSON is critical. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Emphasize Databricks Connect serverless, add setup requirements - Rule #10: Use Databricks Connect Serverless by default (avoid execute_code) - Setup: Use uv, require Python 3.12 and databricks-connect>=16.4 - Common issues: Add DatabricksEnv ImportError and Python version fixes 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix MCP server crash on request cancellation When a client cancels a long-running MCP request, there's a race condition between the cancellation and normal response paths: 1. Client cancels request → RequestResponder.cancel() sends error response and sets _completed = True 2. Middleware catches CancelledError and returns a ToolResult 3. MCP SDK tries to call message.respond(response) 4. Crash: assert not self._completed fails Fix: Re-raise CancelledError instead of returning a result, allowing the MCP SDK's cancellation handler to properly manage the response lifecycle. See: modelcontextprotocol/python-sdk#1153 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Add structured_content to error responses for MCP SDK validation When tools have an outputSchema (auto-generated from return type like Dict[str, Any]), MCP SDK requires structured_content in all responses. The middleware was returning ToolResult without structured_content for error cases (timeout, exceptions), causing validation errors: "Output validation error: outputSchema defined but no structured output returned" Fix: Include structured_content with the same error data in all error responses. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Migrate KA operations to Python SDK and fix name lookup issues - Migrate ka_create, ka_get, ka_sync_sources to use Python SDK - Keep ka_update using raw API 2.1 due to SDK FieldMask bug (converts snake_case to camelCase but API expects snake_case) - Fix find_by_name to sanitize names (spaces→underscores) before lookup - Fix ka_create_or_update to lookup by name when no tile_id provided, preventing ALREADY_EXISTS errors on repeated calls - Update MCP tool layer to use new flat response format - Map SDK state values (ACTIVE, CREATING, FAILED) to endpoint_status - Add integration test for updating existing KA via create_or_update 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix knowledge source description requirement and test ordering - Provide default description for knowledge sources when not specified (API requires non-empty knowledge_source.description) - Move KA update test to after endpoint is ONLINE (update requires ACTIVE state) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix structured_content not populated for tools with return type annotations FastMCP auto-generates outputSchema from return type annotations (e.g., -> Dict[str, Any]) but doesn't populate structured_content in ToolResult. MCP SDK validation then fails: "outputSchema defined but no structured output" Fix: Intercept successful results and populate structured_content from JSON text content when missing. Only modifies results when: 1. structured_content is missing 2. There's exactly one TextContent item 3. The text is valid JSON that parses to a dict 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * fix(mcp): apply async wrapper on all platforms to prevent cancellation crashes The asyncio.to_thread() wrapper was only applied on Windows, but it's needed on ALL platforms to enable proper cancellation handling. Without this fix, when a sync tool runs longer than the client timeout: 1. Client sends cancellation 2. Sync tool blocks event loop, can't receive CancelledError 3. Tool eventually returns, but MCP SDK already responded to cancel 4. AssertionError: "Request already responded to" → server crashes This was discovered when uploading 7,375 files triggered a timeout, crashing the MCP server on macOS. Extends the fix from PR #411 which added CancelledError handling in middleware - that fix only works when cancellation can propagate, which requires async execution via to_thread(). 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Fix: don't set structured_content on error responses Setting structured_content causes MCP SDK to validate it against the tool's outputSchema. For error responses, the error dict {"error": True, ...} doesn't match the expected return type (e.g., Union[str, List[Dict]]), causing "Output validation error: 'result' is a required property". Fix: Only set structured_content for successful responses, not errors. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * Improve dashboard skill structure based on error analysis - Add JSON skeleton section to SKILL.md showing required structure - Add Genie note clarifying it's not a widget (use genie_space_id param) - Move Key Patterns to top of 4-examples.md for discoverability - Clarify example is reference only - adapt to user's actual requirements - Add structural errors table to 5-troubleshooting.md Root cause fixes: - queryLines must be array, not "query": "string" - Widgets must be inline in layout[].widget, not separate array - pageType required on every page 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Quentin Ambard <quentin.ambard@databricks.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: calreynolds <calrey98@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
When clients cancel long-running tool calls, the server crashes with an assertion error and becomes unresponsive. This PR fixes the issue by preventing duplicate responses.
Problem
Solution
Catch CancelledError in _handle_request and return early without sending a second response.
Testing
Added test that cancels a tool call and verifies the server remains responsive.
Fixes #1152
Related: PrefectHQ/fastmcp#508, PrefectHQ/fastmcp#823