cartograph-mcp

mcp-name: io.github.benteigland11/cartograph

An MCP server for Cartograph that exposes the daily widget workflow for agents without mirroring the entire CLI. Search, inspect, install, create, validate, check in, and configure Cartograph defaults through a compact agent-facing surface, then fall back to the CLI for the full administrative and recovery surface.

Why this exists

The Cartograph CLI is the source of truth, but agents do better when the common path is small and explicit.

This MCP keeps the top-level tool surface focused on daily driving:

• finding reusable widgets
• inspecting and installing them
• managing installed widget copies
• creating new widgets
• validating and checking them back in
• adjusting the core Cartograph defaults that affect normal workflow

Everything else stays in the CLI. That keeps the MCP easier to teach, easier to test, and less likely to drift into a second full interface.

Quick start

pip install cartograph-mcp

Claude Desktop example:

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

The package depends on cartograph-cli and shells out to it as the source of truth for the full command surface.

Common CLI setup commands:

# Claude Code
claude mcp add cartograph --scope user -- cartograph-mcp

# Codex
codex mcp add cartograph -- cartograph-mcp

# Gemini CLI
gemini mcp add cartograph cartograph-mcp

# Cursor
cursor --add-mcp '{"name":"cartograph","command":"cartograph-mcp"}'

Claude Code expects an explicit scope flag such as --scope user.

Tool surface

The MCP intentionally exposes a small workflow-oriented surface:

• registry_widget Actions: search, inspect, install, rate
• installed_widget Actions: upgrade, uninstall
• widget_status
• create_widget
• validate_widget
• checkin_widget
• cartograph_config
• cartograph_rules

These are not a 1:1 mirror of the CLI. They are grouped around agent intent:

• registry-facing work
• installed-widget mutation
• project health/status
• widget authoring
• workflow configuration
• custom validation rules

Example workflow

1. Search the registry before writing logic.
2. Inspect the widget you want to reuse.
3. Install it into the project.
4. If no existing widget fits, create one.
5. Validate it with the full dry-run pipeline.
6. Check it in with a reason once it is ready.

In Cartograph terms:

• registry_widget handles discovery and install
• installed_widget handles already-installed widget paths like cg/backend_retry_python
• validate_widget is the dry run for checkin_widget
• cartograph_config manages the defaults that change how your day-to-day loop behaves
• cartograph_rules manages custom rules that run during validate and checkin

Philosophy

This MCP is deliberately not the whole CLI.

The common path belongs in MCP. The official full surface belongs in cartograph.

For uncommon, administrative, or recovery operations, use:

cartograph --help
cartograph <command> --help

That includes things like rollback/delete, cloud operations, auth, setup, rules, doctor, export/import, and other non-daily commands.

Configuration

cartograph_config exposes the workflow defaults that matter most to agents:

• auto-publish
• visibility
• governance
• cloud
• show-unavailable
• publish-registry

Reading and writing config is done through the CLI's --json path so MCP can consume it safely.

Testing

This package is tested in two layers:

• command-contract tests that mock the CLI runner and assert the exact commands the MCP buil

... View full README on GitHub