Per-tool timeout overrides

[dgenio]

Description

Context

https://github.com/modelcontextprotocol/python-sdk/pull/1685 implemented server-wide tool execution timeouts with a default of 300 seconds.

Motivation

Some tools may need different timeout values:

• Quick lookups: get_weather() - 10 seconds
• Heavy processing: analyze_dataset() - 20 minutes

Proposed API

# Server default
mcp = FastMCP("MyServer", tool_timeout_seconds=300)

# Per-tool override
@mcp.tool(timeout_seconds=1200)  # 20 minutes
async def analyze_large_dataset(data: str) -> dict:
    ...

Questions

• Is this needed in practice, or is the server-wide timeout sufficient?
• Should None mean "use server default" or "no timeout"?
• Should this be per-tool or per-call?

Non-Goals (for this issue)

• Client-side timeout control (separate concern)
• Retry policies (separate concern)

References

No response