MCP server maintainers have no visibility into how LLMs experience their tools. We're proposing a protocol-level mechanism where the LLM analyzes its own session and reports structured, anonymized feedback — tool confusion, reliability issues, documentation gaps, missing capabilities — with zero user effort after opt-in. SEP draft: https://gist.github.com/dkindlund/79223feb9f0ecd5fe2dbc6abf5... TypeScript prototype (10 tests): https://github.com/dkindlund/typescript-sdk/tree/feat/client... |
No comments yet