MCP tool overview

The MCP server (@matthesketh/second-brain-mcp) exposes six tools. All are backend-agnostic — they operate against whichever BrainAdapter you configured.

Tool	Purpose
brain_setup_check	Diagnose what’s installed/missing in this Claude instance
brain_seed_taxonomy	Create the root Claude Memory note + standard children
brain_recall	Search the brain for notes relevant to a query
brain_remember	Write a new note with category enforcement
brain_update	Update a note with archived prior content + reason
brain_scan_transcripts	Mine ~/.claude/projects/* for inferable Standards (suggests only — never auto-writes)

Discipline contract

The plugin’s _discipline skill enforces several rules around how these tools should be used:

1. Cite when consulting — every reference to a brain_recall result includes the note title + last-modified date
2. Verify before asserting — facts from the brain are checked against current code/state before action
3. Decisions need what / why / evidence — brain_remember refuses to write a decision/lesson missing any of the three
4. Never silently overwrite — brain_update archives prior content as a child note before replacing
5. Surface contradictions — if the brain says X but live state shows Y, raise it

These aren’t tool-level enforcement (the MCP doesn’t know what Claude says in chat); they’re skill-level. See Discipline rules for the full set.