claude-code-hooks-mcp-server

MCP server that lets Claude Code inspect and edit its own hooks in ~/.claude/settings.json. Meta — Claude managing the rules it runs under.

Tools (read-only by design)

| Tool | What it does | | -------------------------- | -------------------------------------------------------- | | claude_hooks_list_all | All hooks across all events, flat | | claude_hooks_for_event | Hooks for one event (PreToolUse, PostToolUse, Stop, ...) | | claude_hooks_list_events | Valid event names Claude Code recognizes |

Why read-only?

Editing ~/.claude/settings.json from inside a Claude session is self-modification of the agent's own security controls — including the destructive-bash guard hook that exists to keep the agent honest. The MCP intentionally exposes read-only inspection only. To edit hooks: open ~/.claude/settings.json directly, or use Claude Code's /hooks slash command.

A future scope expansion could add edit tools behind an explicit --unsafe flag opted in at server-launch time. Not in scope today.

Setup

pnpm install
pnpm build
pnpm smoke   # reads your real settings.json, prints a summary

Add to ~/.claude.json:

"mcpServers": {
  "hooks": {
    "command": "node",
    "args": ["/absolute/path/to/claude-code-hooks-mcp/dist/index.js"]
  }
}

Restart Claude Code.

Optional override: set CLAUDE_SETTINGS_PATH in the MCP env to point at a different settings file (e.g. a project-scoped .claude/settings.json).

Pairs with

• claude-code-hook-eval skill — measure each hook's wall-clock saved/spent. Combine: claude_hooks_list_all to enumerate, then run the eval against the list, then claude_hooks_remove for losers.

License

MIT.