integrations

Claude Code Integration

Install Docmancer skills, project-aware recall, separate local capture, and writable MCP for Claude Code.

Updated

Install the skills

docmancer install claude-code

This installs the docs skill and memory skill under ~/.claude/skills/, then adds a managed recall instruction to ~/.claude/CLAUDE.md. The installed guidance teaches Claude Code to query project-aware memory, add a durable record only when the user explicitly asks, inspect before changing memory, and avoid automatic team writes.

Use --project to install the skill under the current repository instead.

Add read-only recall hooks

docmancer install claude-code --hooks

Recall hooks use SessionStart and UserPromptSubmit. They pass the current working directory so matching project and team memory can join global context while unrelated repositories stay out. Weak matches, malformed events, timeouts, and stale indexes produce no output and cannot block a turn.

Remove recall without changing capture:

docmancer remove claude-code --hooks

Opt into capture separately

docmancer install claude-code --capture-hooks

Claude Code capture uses PostCompact and SessionEnd. It redacts a bounded local payload or transcript tail, extracts durable memory atoms, and stores only those Markdown records. It never copies a raw transcript, calls a hosted model, or promotes directly into team memory.

Remove capture without changing recall:

docmancer remove claude-code --capture-hooks

Install the local MCP

pipx inject docmancer mcp
docmancer mcp install claude-code --yes

The stdio server exposes memory search, add, list, show, forget, and promote plus docs search. Forget and promotion require an explicit confirmation parameter. The MCP server is local and does not require a hosted Docmancer account.

Verify

docmancer doctor
docmancer memory status
docmancer memory query "what did we decide?" --project "$PWD"