Refine's MCP Server: Put Your GEO on Autopilot with Any AI Assistant

Summary

Refine now exposes its entire GEO platform through a remote MCP (Model Context Protocol) server. Any MCP-compatible AI assistant — Claude Desktop, Claude Code, ChatGPT, Cursor, open-source agents — can read your AI visibility data, manage your monitored prompts and competitors, and generate publish-ready content, all from a conversation. This post explains what the server exposes, how to connect in two minutes, and the workflows it unlocks.

What Is MCP, and Why It Matters for GEO

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic and since adopted across the industry, that gives AI assistants a uniform way to call external tools and read external data. Instead of every product building a custom plugin for every assistant, a product ships one MCP server and every MCP client — Claude, ChatGPT, Cursor, Windsurf, open-source agent frameworks — can use it.

For GEO (Generative Engine Optimization), this closes a loop that used to be manual. Your visibility data lives in a dashboard; your content work happens in an AI assistant; the strategy connecting them lives in your head. With Refine's MCP server, the assistant that writes your content can also see your mention rate, your average rank, your competitor share and the raw AI answers behind them — and act on what it finds.

What the Refine MCP Server Exposes

The server is a standard remote Streamable HTTP MCP server. It exposes fifteen tools across five groups, mirroring what you can do in the Refine dashboard:

• Analytics — list your monitored prompts, get your current visibility summary (mention rate, average rank), pull per-day visibility history filtered by prompt or model, and read the raw AI-search test results behind any data point.
• Prompts — create, update and delete the search prompts Refine monitors for your brand.
• Competitors — add or remove the competitors you benchmark against, with their ranking and visibility share.
• Tags — organize prompts with color-coded tags.
• Content generation — kick off asynchronous generation of blog articles, Reddit responses or LinkedIn posts from any monitored prompt, then poll for the finished text.

Connect in Two Minutes

Three steps: create an API key from your Refine dashboard, point your client at the endpoint, and ask a question. Keys start with rfn_live_ and are shown once at creation. With Claude Code, connecting is a single command:

claude mcp add --transport http refine \
  https://app.refine-app.com/api/mcp/mcp \
  --header "Authorization: Bearer rfn_live_your_key"

Claude Desktop, Cursor and most other clients take a small JSON config with the same endpoint and header. We publish copy-paste snippets for every major client in the docs, plus dedicated setup guides for Claude, OpenAI and open-source agents on this blog.

What Your Agent Can Do Once Connected

The point of MCP is not individual API calls — it is that the assistant chains them. Real workflows our early users run:

• "Which of my prompts lost visibility this month, and why?" — the agent pulls 30-day history, isolates the drops, reads the raw AI answers, and reports which competitor took the mentions.
• "Draft a blog article targeting my weakest prompt." — the agent finds the prompt with the lowest mention rate, calls generate_content, polls until the article is ready, and hands you the draft.
• "Rebuild my prompt list around how buyers actually search." — the agent reviews your current prompts, proposes a better universe, and creates, updates and tags them in one pass.
• "Give me a Monday-morning GEO report." — on a schedule, an agent summarizes mention rate, rank movement and competitor share, and posts it wherever your team lives.

Security, Limits and Plans

• Each API key is scoped to a single brand and shown only once at creation. You can hold up to 20 active keys and revoke any of them at any time.
• Requests are rate-limited to 60 per minute per key, with a 300-second cap per request.
• Analytics, prompt, tag and competitor tools are available on the Visibility plan and above. Content generation requires the Autopilot plan.
• Content-generation source URLs are validated server-side — only public http/https URLs are accepted.

Where This Is Going

AI search changed how buyers discover brands; agents are changing how teams operate tools. The two trends meet exactly here: the same assistants your buyers use to find products are the ones your team can now use to win those recommendations. Connecting your GEO stack to your agents is not a novelty — it is what "GEO on autopilot" actually looks like in practice.