MCP tool listing hangs indefinitely when server is unresponsive
@thiagob 已经在做这个了。
开始于 2026年9月14日。
- 主要语言
- Python
- 星标
- 28
- 派生
- 42
- 平均合并
- 2 天 13 小时
- 30 天内合并 PR
- 35
描述
Describe the Bug
Issue
When listing MCP tools, if the MCP server is unresponsive or is throwing some error, session.initialize()
gets stuck indefinitely/keeps retrying instead of failing fast.
Likely cause: session.initialize() in /agentgateway/_lob.py (lines 350
and 482) has no deadline at the MCP protocol level. While self._config.timeout
(default 60 s) is passed down to httpx.AsyncClient, it only bounds individual
HTTP chunk reads — SSE keep-alives from the server continuously reset this timer,
so the call never actually times out.
Steps to Reproduce
Try listing tools from an MCP server that isn't working or is unreachable. The AGW client keeps retrying until the upstream request times out
Expected Behavior
It should retry once or twice and then error out
Screenshots
No response
Used Versions
https://github.tools.sap/agent-extensibility/extensibility-agent/blob/main/requirements.txt
Code Examples
# Your code here
Stack Trace
No response
Log File
Log file
...Affected Development Phase
Development
Impact
Blocked
Timeline
This is a requirement for EAC
贡献指南
从这里开始
- 先读完整个 Issue,再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库,在一个分支上完成修改。
- 提交 Pull Request,并在描述里引用这个 Issue 编号。
评估
这个 Issue 还没有评估数据。