brain-mcp

Experimental MCP server wrapping brain-cli's brain.py (personal knowledge DB at ~/brain.db, or $BRAIN_DB) as 6 MCP tools: brain_save, brain_search, brain_get, brain_context, brain_link, brain_feedback. Built with FastMCP.

brain.py itself is not modified. All adaptation (stdout-capture guard, Namespace synthesis, error shaping) lives in brain_mcp/adapter.py and brain_mcp/server.py.

Status: local prototype, both transports smoke-tested manually. No auth, no deployment yet — see brain-cli for the underlying CLI and schema.

Setup

Clone this repo next to brain-cli (the pyproject.toml path-depends on ../brain-cli):

git clone https://github.com/victorsabino/brain-cli.git
git clone https://github.com/victorsabino/brain-mcp.git
cd brain-mcp && uv sync

Related MCP server: Hoard

Run

# stdio (default)
uv run --directory . brain-mcp
# or, for the inspector:
fastmcp dev brain_mcp/server.py

# streamable-http
BRAIN_MCP_TRANSPORT=http BRAIN_MCP_HOST=127.0.0.1 BRAIN_MCP_PORT=8000 \
  uv run --directory . brain-mcp
# serves at http://127.0.0.1:8000/mcp

BRAIN_DB overrides the default ~/brain.db — set it before starting the server.

Out of scope (CLI-only, not exposed as MCP tools)

migrate, reindex, consolidate, delete, update, invalidate, harvest, review, reconcile, anchor, artifact, secrets, block, doctor, stats, recent, history, tags.