Connecting Rankmio to Claude, Cursor and other AI agents via MCP

Why a coding agent without this data optimises past real demand, and what that has to do with vibe coding, is explained in our article Why Claude Code only reaches its full potential with SEO and GEO context.

MCP connects an AI agent to outside data

The Model Context Protocol (MCP) is an open standard that lets AI programs call the tools of other services. The agent receives a list of tools with descriptions from Rankmio, such as “read the open tasks of a project” or “start a GEO audit”. It decides for itself which tool it needs for your question, calls it and works with the answer.

So you no longer ask “Read this CSV file”, but “Which pages of rankmio.com lost the most clicks last month, and why?”. The agent fetches the figures itself.

The API key determines what the agent may do

Every connection needs an API key. You create it in your Rankmio account in the “API keys” tab and choose one of three levels of rights.

With “Read only” the agent sees your data and changes nothing. It reads, for example, the visibility index, open tasks, click losses, AI citations, backlinks and the Google index status of each page. I recommend this level to start with.

With “Read and act” the agent may also tick off or hide tasks, save keywords, sync Search Console, report changed pages to Bing via IndexNow, create briefs and send drafts to WordPress. It asks you before every transfer. Before it overwrites an existing post, it shows you what changes.

With “Read, act and spend credits” the agent may also generate drafts and start research, GEO audits, citation scans and crawls. Rankmio requires a confirmation for this: The first call only returns the price, and only a second call with that price books it. So the agent has to ask you before credits are spent.

Whatever the level, the agent may do nothing your role does not allow in the web interface. Functions from packages you have not booked stay locked for it too. It cannot delete anything.

Rankmio sets fixed limits for every key

A key spends at most 200 credits a day. It may make 60 calls per minute and 1,000 per day. An agent starts a crawl of the website at most every 30 minutes per project and fetches backlink data at most every seven days. A crawl puts load on the website, and backlinks change slowly.

A ticked-off task reopens if the next measurement still finds the problem. So an agent cannot simply make work disappear.

Step 1: Create the API key

1. Log in to Rankmio and open your account.
2. Switch to the “API keys” tab.
3. Give the key a name you will recognise later, such as “Claude Code laptop”.
4. Choose the rights and create the key.
5. Copy the key straight away. It starts with rmk_ and is shown in full only this once.

Create a separate key for every program and every device. If a device gets lost, you revoke only that one key.

Step 2: Add Rankmio to your AI program

All programs need the same two details. The address is https://rankmio.com/mcp, and the header Authorization: Bearer rmk_… with your key goes with it. If you use Rankmio in German, take https://rankmio.de/mcp. The programs only differ in where these details go.

The configuration files may change with new versions of the programs. If an example does not work, the program’s own documentation applies.

Claude Code

In Claude Code one command in the terminal is enough:

claude mcp add --transport http rankmio https://rankmio.com/mcp --header "Authorization: Bearer rmk_…"

claude mcp list checks whether the connection works. Inside a running Claude Code session, /mcp shows the status and the tools.

Cursor

Add Rankmio to the file ~/.cursor/mcp.json, or for a single project to .cursor/mcp.json in the project folder:

{
  "mcpServers": {
    "rankmio": {
      "url": "https://rankmio.com/mcp",
      "headers": { "Authorization": "Bearer rmk_…" }
    }
  }
}

Under Cursor Settings → MCP you then see whether Rankmio is connected.

Claude Desktop

Claude Desktop connects servers with a fixed key through the small helper program mcp-remote. Node.js has to be installed for this. In Claude Desktop, open Settings → Developer → Edit Config and add:

{
  "mcpServers": {
    "rankmio": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://rankmio.com/mcp", "--header", "Authorization:${RANKMIO_AUTH}"],
      "env": { "RANKMIO_AUTH": "Bearer rmk_…" }
    }
  }
}

Then quit Claude Desktop completely and start it again. The detour via env is deliberate: On Windows, Claude Desktop splits spaces in the arguments incorrectly, so the value with the space goes into the environment variable.

VS Code with GitHub Copilot

Create the file .vscode/mcp.json in the project:

{
  "servers": {
    "rankmio": {
      "type": "http",
      "url": "https://rankmio.com/mcp",
      "headers": { "Authorization": "Bearer rmk_…" }
    }
  }
}

In the Copilot chat, choose agent mode. Rankmio then appears in the list of tools.

Windsurf

Windsurf expects the address under serverUrl, not under url. The file is ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "rankmio": {
      "serverUrl": "https://rankmio.com/mcp",
      "headers": { "Authorization": "Bearer rmk_…" }
    }
  }
}

Windsurf does not reload the file by itself. Click “Refresh” in the MCP view afterwards.

Gemini CLI

In ~/.gemini/settings.json the address goes under httpUrl:

{
  "mcpServers": {
    "rankmio": {
      "httpUrl": "https://rankmio.com/mcp",
      "headers": { "Authorization": "Bearer rmk_…" }
    }
  }
}

/mcp inside a running Gemini CLI session shows whether Rankmio is connected.

Step 3: Ask the first question

Start with a question that only reads. That way you see whether the connection works without anything happening. The agent first calls list_projects and then knows your projects.

Good first questions are, for example:

• “Show me my Rankmio projects and the visibility index of each.”
• “Which five open tasks of rankmio.com matter most, and what exactly needs to be done for each?”
• “Which pages lost the most clicks in the last 28 days?”
• “What does a GEO audit cost for my project?” The answer costs nothing, the tool get_prices only reads.

A typical workflow with a coding agent

With Claude Code or Cursor in your website’s folder, the data turns into work on the code. The agent reads the open tasks, changes titles, descriptions or text in your repository and deploys the change. It then reports the changed pages to Bing via IndexNow and uses recheck_task to check whether the task is done.

For WordPress websites the same route works without a repository. The agent loads an existing post into the Content Studio editor, revises it and writes it back after your confirmation. How the connection to WordPress works is described in our article Sending GEO content to WordPress without installing a plugin.

Frequently asked questions