展示HN:PolyMCP – 开源工具包,用于暴露MCP工具并运行代理
嗨,HN——我是PolyMCP(MIT)的作者。
PolyMCP是一个用于模型上下文协议(MCP)的工具包/运行时:它帮助您构建MCP工具服务器(通过HTTP/标准输入输出将Python函数暴露为工具),连接多个MCP服务器,并运行能够协调它们的代理,提供便于部署的保护措施。
最新的更新主要集中在可靠性方面:
- OAuth2(RFC 6749):客户端凭据 + 授权码流程,刷新处理,基本重试
- Docker执行器清理修复(Windows + Unix)
- 更好的技能/工具匹配 + 标准输入输出服务器支持
- CodeAgent异步处理 + 错误恢复
- 健康检查、结构化日志记录和速率限制
代码库:https://github.com/poly-mcp/Polymcp
如果您已经部署了MCP/代理系统:最常遇到的问题是什么(认证、速率限制、可观察性、沙箱)?
我特别感兴趣的是OAuth2的边缘案例和容器生命周期问题。
查看原文
Hi HN — I’m the author of PolyMCP (MIT).<p>PolyMCP is a toolkit/runtime for the Model Context Protocol (MCP): it helps you build MCP tool servers (expose Python functions as tools over HTTP/stdio), connect multiple MCP servers, and run agents that orchestrate them with deploy-friendly guardrails.<p>Latest update is mostly reliability work:
- OAuth2 (RFC 6749): client credentials + authorization code flows, refresh handling, basic retries
- Docker executor cleanup fixes (Windows + Unix)
- Better skills/tool matching + stdio server support
- CodeAgent async handling + error recovery
- Health checks, structured logging, and rate limiting<p>Repo: https://github.com/poly-mcp/Polymcp<p>If you’ve deployed MCP/agent systems: what tends to bite you first (auth, rate limits, observability, sandboxing)?
I’m especially interested in OAuth2 edge cases and container lifecycle issues.