Mealie MCP Server

Docker-hosted MCP server for Mealie. It gives an AI agent the tools to read a cookbook (e.g. a PDF), extract the recipes (translating them if needed) and store them in Mealie — then verify its own work.

Built against Mealie API v3.19.2. Transport: SSE (http://localhost:8001/sse) — persistent container, no subprocess spawning.

It deliberately covers only the endpoints that serve the cookbook → Mealie workflow (recipes, organizers, foods, units) plus read endpoints for verification. The rest of the Mealie API is intentionally omitted.

Prerequisites

• Docker (with Compose)

• A reachable Mealie instance

• A Mealie API token (or a username/password)

Related MCP server: HowToCook-MCP Server

Generate an API token

In Mealie: avatar → Settings → API Tokens → Create (or /user/profile/api-tokens). Copy the token — it is shown only once.

Quick Start

git clone https://github.com/NigelVanHattum/mealie-mcp.git
cd mealie-mcp

# Configure credentials
mkdir -p ~/.config/mealie-mcp
cat > ~/.config/mealie-mcp/config.json << 'EOF'
{
  "base_url": "https://mealie.example.com",
  "api_token": "your-api-token-here",
  "verify_ssl": true
}
EOF

# Start the server
docker compose up -d

The server runs at http://localhost:8001. Verify:

curl http://localhost:8001/health
# {"status": "ok"}

Connect to Claude

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "mealie": {
      "url": "http://localhost:8001/sse"
    }
  }
}

Restart Claude Desktop, then try: "Check my Mealie connection" → it calls get_current_user.

Claude Code

claude mcp add mealie --url http://localhost:8001/sse

Configuration

Loaded in this order (first found wins):

The server authenticates to Mealie with API keys only — there is no username/password path.

Exposure & security

The MCP API itself is unauthenticated: /sse, /messages/ and /health have no inbound auth. Anyone who can reach the port can call every tool — including create_recipe, overwrite_recipe and delete_recipe — using the server's Mealie API key.

So treat the server as a private backend: let Obot (or your MCP host) connect to it over localhost or a private/Docker network, and do not expose the port to the public internet. If it must traverse an untrusted network, put a reverse proxy with authentication in front of it. Transport is SSE; Streamable HTTP can be added if your host prefers it.

Recommended agent workflow

1. get_current_user — confirm the token works and note the active group.

2. list_recipes with search — check the recipe doesn't already exist.

3. create_recipe — one call stores a full recipe. Translate fields to the target language (e.g. Dutch) before calling. Pass ingredients and instructions as plain strings; pass categories / tags / tools as names (they are created automatically if missing).

4. get_recipe with the returned slug — verify the stored content.

create_recipe and update_recipe merge onto the recipe's current state, so they never wipe fields you didn't provide. When you want a clean re-import, use overwrite_recipe instead: it replaces all recipe content explicitly — every content field you omit is cleared (identity, settings and image are kept).

Tools Reference

create_recipe — key fields

name (required), description, ingredients (string[]), instructions (string[]), recipeYield, servings, prepTime, cookTime, totalTime, categories (string[]), tags (string[]), tools (string[]), nutrition (object), orgURL, extras (object).

Ingredients are stored as free text by default — ideal for cookbook imports. For structured ingredients, pass an object per line ({quantity, unit, food, note}) and manage foods / units with their tools.

Development

pip install -e ".[dev]"
pytest

# Build image locally (tests run in the build stage — build fails if tests fail)
docker build -t mealie-mcp .

docker compose up

API Reference

Interactive docs for your instance: https://<your-mealie>/docs · OpenAPI: https://<your-mealie>/openapi.json