Concurrent runs sharing an MCP server fail with a cancel scope RuntimeError

Running several agent runs or evals at once against the same MCP server toolset eventually fails with RuntimeError: Attempted to exit cancel scope in a different task than it was entered in.

Difficulty

Easy

Time

~5 min

Steps

Fixed in

v1.92.0

Last checked

Oct 17, 2026

Verified fix: PR #4514 merged on 2026-05-07. Fixed in v1.92.0 (first stable release containing the merge commit). Last checked 2026-10-17.

Symptom

Running several agent runs or evals at once against the same MCP server toolset eventually fails with RuntimeError: Attempted to exit cancel scope in a different task than it was entered in.

Evidence what you'll see

Shows up with max_concurrency above 1 and about 10 or more eval cases
Only with toolsets created from MCP servers
Reported on version 1.0.1

Cause

Each concurrent run entered and exited the shared MCP server with async with, so the session's cancel scope could be exited in a different task from the one that entered it.

Fix steps

0 of 2 done

Step 1Upgrade Pydantic AI to a release that includes PR #4514, which runs each MCP server's session in its own dedicated background task.
Step 2Until then, run cases one at a time with max_concurrency=1; the report only saw the error at higher concurrency.

Your check marks stay in this browser only.

Verify the fix worked

Run the eval with max_concurrency=4 and confirm every case finishes without the RuntimeError.

Sources and credit

Summarized in our own words from these public reports. Credit goes to their authors.

Provided as is, without warranty. Try changes in a non-production environment and back up first. Security-related steps: verify against the official docs. Terms · Request removal or attribution change

Hit an error we don't cover? Submit it