MCP setup

Connect IngestMD to your AI editor

IngestMD ships an MCP server. Point your editor at it and the AI can pull this project as clean, area-scoped context, all on your machine. On macOS the app can do this for you in Settings, Connect to your editor. The manual steps below cover every editor and Windows.

First, get the binary on your machine

Install IngestMD, then open Settings, Command-line tools, and Install command-line tools. That puts ingestmd-mcp at /usr/local/bin/ingestmd-mcp on macOS. The snippets below use that path; change it if you installed elsewhere.

Claude Code

Run this once in a terminal (user scope):

claude mcp add -s user ingestmd -- /usr/local/bin/ingestmd-mcp

After it is added, /mcp in Claude Code lists ingestmd.

Cursor

Add to ~/.cursor/mcp.json (merge with any existing servers):

{
  "mcpServers": {
    "ingestmd": { "command": "/usr/local/bin/ingestmd-mcp" }
  }
}

VS Code

Add to your user mcp.json (~/Library/Application Support/Code/User/mcp.json on macOS):

{
  "servers": {
    "ingestmd": { "command": "/usr/local/bin/ingestmd-mcp", "type": "stdio" }
  }
}

VS Code uses "servers" and needs "type": "stdio".

Codex

Add to ~/.codex/config.toml:

[mcp_servers.ingestmd]
command = "/usr/local/bin/ingestmd-mcp"

Windows

Use the same config shapes, but point command at the Windows binary, ingestmd-mcp.exe inside your IngestMD install folder, for example %LOCALAPPDATA%\\IngestMD\\ingestmd-mcp.exe. The editor config files live under %USERPROFILE% and %APPDATA% (for VS Code, %APPDATA%\\Code\\User\\mcp.json).

The tools

Free for everyone:

  • bundle - Turn a whole repo into Markdown bundles, secrets redacted.
  • discover_areas - Split the repo into its coherent areas and return one bundle per area.
  • list_areas - A menu of the areas: name, file count, languages, approximate token cost. Reads no file contents, so use it first to choose one.
  • area_context - Load just one area as a bundle. Use after list_areas to work on a single subsystem instead of the whole repo.
  • enhance_areas - Area naming runs in the IngestMD desktop app, where you pick the model. Over MCP this returns that guidance; once you enhance in the app, list_areas and area_context show the names automatically.

Pro (require a license):

  • task_context - Assemble a token-budgeted context for a specific task or query: the relevant files plus their tests and dependents.
  • diff_context - A review-ready bundle built from your current git changes.
  • collection - Load a curated collection you saved in the app, by name.

How to prompt it

Ask in plain language and the AI picks the tool. The area workflow is the point: see the parts with list_areas, then load just one with area_context, so the model works on a single subsystem instead of drowning in the whole repo.

  • "List the areas in /path/to/repo." → list_areas
  • "Load only the Auth area from /path/to/repo and answer from it. Ask me before loading any other area." → area_context, scoped
  • "Assemble task context for "fix the doubled blank rows" in /path/to/repo, 50k token budget." → task_context
  • "Build a review bundle of my changes against main in /path/to/repo." → diff_context
  • "Load my saved collection "onboarding"." → collection

For a tight boundary, tell the AI to answer only from the loaded area and to ask before loading another. The app's Areas tab can also copy a ready-made, boundary-aware prompt for a single area. Pro tools need a license; the free tools work for everyone, and everything runs locally with the same secret redaction as the app.

See the use cases for the bigger picture, or download IngestMD to get started.