Temporal integration in `pydantic-ai-slim` unconditionally requires `mcp`

[jeroenvandermerwe]

Initial Checks

• I confirm that I'm using the latest version of Pydantic AI
• I confirm that I searched for my issue in https://github.com/pydantic/pydantic-ai/issues before opening this issue

Description

Importing the Temporal integration (pydantic_ai.durable_exec.temporal, e.g. to use PydanticAIPlugin) unconditionally imports pydantic_ai.mcp.MCPServer, which raises an ImportError unless the external mcp packages are installed. This makes the optional mcp dependencies effectively mandatory for anyone using Temporal, even when no MCP toolsets are used, undermining the purpose of the “slim” variant.

I might be misunderstanding the docs—Is this expected behavior? I’m not using any MCP toolsets; I only import PydanticAIPlugin, and the import fails unless mcp is added. Should mcp be required whenever importing the Temporal module, or only when an MCP toolset is actually used?

Example Code

Python, Pydantic AI & LLM client version

Python: 3.13.7
Pydantic AI (slim, with openai): 1.0.6

[DouweM]

@jeroenvandermerwe Good catch, not intentional! I think #2908 should do it. Can you give that a try?

[jeroenvandermerwe]

Thanks, @DouweM! I tested the PR and the MCP ImportError is resolved. However, importing pydantic_ai.durable_exec.temporal now raises ModuleNotFoundError: No module named 'logfire'. Similar to mcp, logfire is provided as an optional extra, so it probably shouldn’t be required just to import the Temporal module unless Logfire integration is actually used.

[DouweM]

@jeroenvandermerwe Good call, how about now?

[jeroenvandermerwe]

@DouweM Thanks for the quick follow-up and fixes! I’ve pulled the updated PR and can confirm the Logfire import issue is resolved — looks good on my end.