@marchese-md/mcp

MCP server exposing MARCHESE.md tools to Claude Code and any other MCP-compatible client. Same underlying logic as the marchese CLI, exposed as 12 tools over the MCP stdio transport for agents that talk MCP directly instead of shelling out.

Install

Add to your MCP client config (e.g. .mcp.json or claude_desktop_config.json):

{
  "mcpServers": {
    "marchese": {
      "command": "npx",
      "args": ["-y", "@marchese-md/mcp"]
    }
  }
}

Or install globally and point at the binary directly:

npm install -g @marchese-md/mcp

{
  "mcpServers": {
    "marchese": {
      "command": "marchese-mcp"
    }
  }
}

Tools

| Tool | Purpose | |---|---| | marchese_audit_context | Audit session context budget, CLAUDE.md line tax, skill preload, MCP presence | | marchese_validate_spec | Validate MARCHESE.md content against the JSON Schema and semantic rules | | marchese_mvm_route | Resolve the Minimum Viable Model for a task name | | marchese_run_loop | Run a 4-block orchestration loop (trigger, skill, verification, history) | | marchese_log_gotcha | Append a gotcha entry to an in-memory ledger | | marchese_write_history | Format a loop execution result as a JSONL history line | | marchese_ask_user | Format a structured AskUserQuestion block for interactive skills | | marchese_interview_user | Build an interview-me or grill-me prompt with critical questions | | marchese_codex_route | Decide whether to offload a token-heavy task to OpenAI Codex CLI | | marchese_rtk_status | Detect RTK binary and optionally process a PreToolUse payload for bash rewrite | | marchese_improve_system | Run the self-improving improve-system loop | | marchese_knowledge | Knowledge base status or ingest, self-seeded from the repo |

Docs

Full spec: docs/SPEC.md in the main repo.

License

MIT