About canimcp

canimcp is a registry of Model Context Protocol clients. It records what each client supports, the evidence for each claim, and the protocol revision the claim applies to.

Where the data comes from

• Source reads. A client's code is read at a pinned commit, and the entry cites the commit and the file paths.
• The probe. A real client runs against an instrumented fixture MCP server and a mock model, and the record says what the client sent to the model. The captures for every run are in the repository under probe/runs/.
• Telemetry. Counts derived from the Hugging Face dataset evalstate/mcp-clients, at the revision recorded in the data file.
• Declarations. What a client sends in initialize.capabilities, or what a curated list says about it.

No client source is copied here. Entries quote short excerpts with the repository and commit they came from.

Corrections

The registry, the notes and this site are built from one public repository, https://github.com/olaservo/canimcp. To report an error or a client that has changed, open an issue there with the client version and what you saw. A correction that cites a commit or a capture is the quickest to apply.

Licence

The registry and evidence under /data, the research notes and the client pages are licensed under Creative Commons Attribution 4.0 International. The code that builds them is MIT licensed. Telemetry under /data/observed follows the terms of the dataset it is derived from.

For agents

/llms.txt lists every page and data file. Each page is available as Markdown by appending .md to its URL or by sending Accept: text/markdown. The JSON Schemas for the data are served at the URLs in their $id fields, under /schemas.