Ask Recipe for Agents from inside your agent

Add https://recipeforagents.com/mcp-server to any MCP client and your agent can look up which Apify Actor fits a job, read its input and output fields, pull a ready n8n recipe and get the config that installs it. It is free, read-only and needs no sign-in.

Add it to your client

One URL, streamable HTTP, no headers. Pick your client:

Claude Code

claude mcp add --transport http recipe-for-agents https://recipeforagents.com/mcp-server

Most clients list the eight tools as soon as the server is added. If yours asks about authentication, choose none.

What your agent can ask

ToolWhat it answersTry asking
search_toolsFinds the catalogue Actors for a kind of data, with website, price, output fields, required input and a scoped MCP URL.Which tool gets me LinkedIn company headcounts?
get_toolOne Actor in full: example input, fields, files (n8n workflow, SKILL.md, schema) and configs.What input does the Google Maps lead finder need?
find_recipeMatches a task to the 66 n8n recipes (Sheets, Slack, Notion, HubSpot, email and more), use cases and skills.I want new Indeed jobs posted to Slack every morning.
get_recipeA recipe's steps, input, schedule, limits and, on request, the importable n8n workflow JSON.Give me the workflow file for that recipe.
list_categoriesThe nine categories with counts and their main websites.What kinds of data can you find tools for?
list_websitesThe 170 websites covered, each with one MCP URL bundling its main Actors.Which Google products do you cover?
build_mcp_configA ready config for Claude Code, Claude Desktop, Cursor, VS Code, ChatGPT, Codex, Gemini CLI or n8n with the Actors you chose.Add the TikTok tools to my Cursor.
troubleshootSourced fixes for 401, 408, 429, missing tools and other connection errors.Cursor says the Apify server has no tools.

How it fits with Apify's server

This server is the menu, not the kitchen. Your agent asks it what to use; the answer includes a scoped mcp.apify.com URL for the Actors it picked. Add that URL too (the build_mcp_config tool writes the config) and the agent can run them with your Apify account. Every answer links the page on this site it came from, with the date the data was checked.

Technical details

  • Endpoint https://recipeforagents.com/mcp-server, MCP streamable HTTP. A browser opening the URL gets a short description page.
  • Protocol versions 2024-11-05 to 2025-11-25 (with the initialize handshake) and the stateless 2026-07-28 revision (server/discover). No sessions, so any request can retry.
  • Eight read-only tools with input schemas and readOnlyHint; no prompts, resources or sampling.
  • Answers come from the site's published data (412 Actors, 170 websites, 66 recipes), refreshed with every publish.
  • Fair use: about two requests a second per address with short bursts allowed; a busy client gets HTTP 429 and can retry.
  • Registry name com.recipeforagents/recipe-for-agents. Manifest and README: github.com/bolvein-firex/recipe-for-agents-mcp.

Questions

Does this server run scrapers?

No. It answers from the published pages of this site: which Actor to use, how to call it and how to wire it into a workflow. Running an Actor happens on Apify with your own account, through the MCP config or workflow it gives you, and Apify bills that account.

Do I need an account or an API key?

Not for this server: it is read-only and open. The Apify Actors it points to need an Apify account when you run them.

How fresh is the data?

The server reads the same data file the site publishes, so it changes when the site does: new and removed Actors show up after the next publish, usually within the hour.

What do you log?

The web server keeps its usual access log with shortened IP addresses for 14 days. The MCP server itself stores only a count of calls per tool per day, never what you asked.