Teleport's configuration surface is wide enough that knowing what to search for is half the battle, and semantic search over the real pages beats guessing at terminology. The index is baked in, so this is a one-line Docker addition; when the docs move on, rebuilding it is a documented job rather than a mystery.
A search server over the Teleport documentation. The pages from Teleport's own repository are embedded into a Chroma database that ships inside the image, and one tool runs the vector search. The result comes back raw — the calling assistant does the interpreting, and no model runs inside the server.
- Vector search across the Teleport docs pages, returning the matching passages
- The index prepopulated in the repository, so there is no embedding run before first use
- Both transports from the same image: stdio, or SSE with `--sse --host 0.0.0.0`
- A refresh path — copy in newer pages, run `fix_include.py`, then `embed.py` to rebuild the index
- No model inside the server, so nothing is paraphrased before you see it
Docker, and nothing else — `docker run --rm -i stevetelelport/teleport-docs-mcp:v0.1.0` for stdio, or publish port 8000 for SSE. No account and no key. It takes a few seconds to spin up, so expect a pause on the first call rather than an instant answer. Local development uses `uv`. Version 0.1.0.
One command — docker run --rm -i stevetelelport/teleport-docs-mcp:v0.1.0