fix(agentgateway): bound MCP protocol calls with a wall-clock deadline

[wangzhengzhuo05]

What

When an MCP server is unresponsive, session.initialize() / list_tools() / call_tool() in the LoB flow hang indefinitely instead of failing fast.

Root cause

The httpx.AsyncClient timeout only bounds individual HTTP chunk reads. An MCP stream's SSE keep-alives reset that timer continuously, so an unresponsive server never trips the client timeout — the protocol call stays alive forever.

How

Add _mcp_call_with_deadline(), which wraps a protocol call in asyncio.wait_for and raises AgentGatewaySDKError on timeout. Keep-alives cannot extend a wall-clock deadline. Applied to initialize, list_tools (in list_server_tools) and initialize, call_tool (in call_mcp_tool_lob), matching the issue's expectation that listing fails fast with a clear error.

Tests

• test_initialize_timeout_raises_agent_gateway_error — stalled initialize fails fast on list_server_tools.
• test_list_tools_timeout_raises_agent_gateway_error — stalled list_tools fails fast.
• test_call_tool_timeout_raises_agent_gateway_error — stalled call_tool fails fast.
• Mutation-checked: with the deadline removed, the test hangs indefinitely (the exact bug reported).
• All 63 tests/agentgateway/unit/test_lob.py tests pass.

AI-generated code disclosure: This contribution was prepared with the assistance of an AI coding agent (Hermes Agent), following SAP's guideline for AI-generated code contributions. The fix and tests were reviewed and verified locally before submission.

Fixes #313

[cla-assistant]

All committers have signed the CLA.

[wangzhengzhuo05]

Closing this out.

It has been waiting 9 days without any maintainer review, and we are winding down this contribution effort — a stale PR sitting in the queue costs maintainers attention without giving them a decision to make, so it is better closed than left open indefinitely.

If the direction is useful, the branch is intact on the fork (fix/mcp-protocol-deadline, +166/−4 across 2 file(s)) and can be picked up as-is.

What it contains: fix(agentgateway): bound MCP protocol calls with a wall-clock deadline — When an MCP server is unresponsive, session.initialize() / list_tools() / call_tool() in the LoB flow hang indefinitely instead of failing fast.

It was verified locally before opening (target test suite green, plus a mutation check that reverts only the fix and confirms the new regression test fails). Thanks for the time.