Connect a coding agent over MCP

Point an MCP client at the Living Brain endpoint, sign in with a token, and use the five tools it offers.

Living Brain runs an MCP server at

https://mcp.livingbrain.wiki/v1/pages/mcp

It is Streamable HTTP. Any MCP client can connect to it; the commands below are the usual ones for the agents we name on the harnesses page. Which harnesses are tested is listed there: until a harness passes our integration test, it stays marked planned, even when this endpoint works.

Get a token

Either create a personal access token under Settings → API tokens in the web app, or run livingbrain login on your machine, which asks you to approve a code in your browser and returns a token for the workspace you pick.

Connect

claude mcp add --transport http livingbrain https://mcp.livingbrain.wiki/v1/pages/mcp \
  --header "Authorization: Bearer <token>"

Use the same URL and header with codex mcp add, cursor mcp add or opencode mcp add, following each client's own syntax. The token is a secret: do not commit it.

The five tools

Tool	What it does
`brain_search`	Finds pages you may read that hold every word of the query.
`brain_page`	Returns one page's Markdown, by slug.
`brain_context_for`	Returns a short brief for a repository and a task: the pages that mention it, in full, with citations.
`brain_note`	Writes a note to your own private memory, redacted first. It can state temporal facts.
`brain_facts`	Returns what is true about a subject now, or as of a date, with its earlier values.

Every page a tool quotes comes back as a numbered citation, with its title and link, so the agent can cite it.

What the agent can see

The agent sees what its user can see, and nothing more. No tool takes a scope argument. If a page is outside the token's grants, it is reported as not found.

The MCP reference has the discovery metadata and the authorization details.