Connect your AI assistant (MCP)

CiteTrack runs a built-in MCP (Model Context Protocol) server on your own WordPress site, so an AI assistant like Claude or ChatGPT, or an automation tool like n8n, can query your live CiteTrack data in plain language (“which prompts am I losing?”, “show this month’s AI revenue”).

Step 1: Generate your token & URL (in WordPress)

1. Go to CiteTrack AI → API & MCP.
2. Click Generate token and copy it now, the full value is shown only once. (Lost it? Just rotate and use the new one.)
3. Note your MCP endpoint URL on the same screen. It is your site followed by /wp-json/citetrack/v1/mcp, e.g. https://yoursite.com/wp-json/citetrack/v1/mcp.

The server accepts the token two ways, use whichever your client supports:

• Header: Authorization: Bearer YOUR_TOKEN
• URL token: append ?token=YOUR_TOKEN to the endpoint (handy when a tool only lets you paste a URL).

Current security rules

• Use the bearer header when possible. It keeps the token out of browser history, server logs and copied URLs.
• Batch requests are rate-limited. MCP batch calls return valid JSON-RPC errors instead of unbounded work.
• Tokens are tied to the WordPress owner. If the token owner is deleted, demoted or loses the required capability, the token stops working.
• Rotate tokens from API & MCP. Rotation immediately invalidates the old value.

Step 2: Connect Claude

Easiest, remote connector. In Claude, open Settings → Connectors → Add custom connector, name it CiteTrack, and paste your endpoint with the token appended:

https://yoursite.com/wp-json/citetrack/v1/mcp?token=YOUR_TOKEN

Save, then enable the CiteTrack connector in a chat. If the connector form has a separate header/auth field, paste the bare endpoint there and put Authorization: Bearer YOUR_TOKEN in the header instead.

Claude Desktop config file (alternative). Open Claude Desktop → Settings → Developer → Edit config, add the block below, and restart Claude. It uses the mcp-remote bridge (fetched on demand by npx; needs Node.js):

{
  "mcpServers": {
    "citetrack": {
      "command": "npx",
      "args": ["-y", "mcp-remote",
               "https://yoursite.com/wp-json/citetrack/v1/mcp",
               "--header", "Authorization: Bearer YOUR_TOKEN"]
    }
  }
}

Step 3: Connect ChatGPT, n8n or your own code

Any MCP-capable client connects the same way, point it at the endpoint and pass the token. In n8n, use the MCP Client node with the endpoint URL and an Authorization: Bearer YOUR_TOKEN header. In ChatGPT, add it as a custom connector using the same URL and bearer token. From code, POST JSON-RPC (initialize, then tools/list / tools/call) to the endpoint with the bearer header.

What the assistant can do

Once connected it can call these read-only tools: snapshot (everything at a glance), crawl_health, share_of_voice, citations, and reputation_changes. So you can ask “what changed in how AI describes us this week?” or “which pages should I refresh to win back citations?” and it answers from your real data.

Verify & troubleshoot

• “Unauthorized” / token errors, the token is wrong or was rotated. Re-copy it from API & MCP, or generate a fresh one and update your client.
• Connector cannot reach the URL, open the endpoint in a browser; you should get a small JSON error, not a 404. A 404 means permalinks need flushing (Settings → Permalinks → Save) or a security plugin is blocking the REST API.
• Header gets stripped, some hosts drop the Authorization header; use the ?token= URL form instead.

← All articles