MCP Server

LaraPaper can expose an HTTP Model Context Protocol server so AI clients (Cursor, Claude Code, and others) can list, create, update, and render recipe plugins for the authenticated user.

The endpoint is POST /mcp. Native plugins are not available through these tools. Recipes are identified by numeric database ID, not UUID. Markup defaults to Blade; pass markup_language: liquid when writing Liquid templates.

Enable

  1. Sign in as the first user.
  2. Go to Settings → Lab and turn on MCP Server.
  3. Go to Settings → API Tokens, choose token type MCP, create a token, and copy it. The value is shown only once.

A regular API token (* ability) is rejected. The MCP endpoint requires the mcp ability. If the lab flag is off, /mcp returns 404.

You can also set TOGGLE_MCP=true instead of using the Lab UI. See Environment variables.

Connect a client

Point the client at your public LaraPaper URL (APP_URL) plus /mcp, and send the MCP token as a bearer header.

Cursor / Claude Code (mcp.json):

{
  "mcpServers": {
    "larapaper": {
      "url": "https://YOUR_LARAPAPER_HOST/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_MCP_TOKEN"
      }
    }
  }
}

Node-based MCP clients can fail against local HTTPS. Prefer the public https:// origin you already use in the browser.

Tools

Tool Purpose
list-recipes List recipes you own (id, name, browser URL, data strategy).
get-recipe Full details: markup, render settings, data payload.
create-recipe Create a recipe. Optional Blade/Liquid, polling/webhook/static, polling URL. Returns id and editor URL.
update-recipe-markup Update full, half_horizontal, half_vertical, quadrant, or shared markup.
update-recipe-settings Data strategy, framework version, polling. Returns a webhook URL when strategy is webhook.
render-recipe Render a layout size (full by default). Returns HTML, or a compile error to fix.

Typical loop: list-recipes or get-recipe → render-recipe → update-recipe-markup / update-recipe-settings → render again. Use create-recipe when you need a new recipe.

Example

After the server is connected, ask your AI client to build and verify a recipe:

Using the LaraPaper MCP server, create a Blade recipe named “Morning weather” that polls https://api.open-meteo.com/v1/forecast?latitude=48.2&longitude=16.4&current=temperature_2m. Write a TRMNL full layout that shows the temperature. Call render-recipe and fix any errors until it succeeds, then give me the recipe URL.

That creates a polling recipe, writes markup, and iterates on render-recipe until the layout compiles. Open the returned URL in the recipe editor to preview it in the browser.

Previous Experimental features Next Local setup