collimer-mcp

· MIT · Model Context Protocol

Let any AI agent run a free AI-search visibility scan on any website — how visible a brand is in ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews.

Trackers tell you you're invisible. Collimer tells you why, and what to fix.

It exposes one tool, collimer_scan, which returns a depth-gated teaser: an AI-visibility score (0–100), a confidence interval, the single biggest gap, and a branded report URL. The full ranked fix plan and verification re-scan unlock with a free account at collimer.com.

Install

Add Collimer to any MCP client. It runs on-demand via npx — no install step.

Claude Desktop / Claude Code — add to your MCP config (claude_desktop_config.json, or .mcp.json in your project):

{
  "mcpServers": {
    "collimer": {
      "command": "npx",
      "args": ["-y", "collimer-mcp"]
    }
  }
}

Cursor / Windsurf / VS Code — same block, in the editor's MCP settings (~/.cursor/mcp.json, Windsurf MCP settings, or .vscode/mcp.json).

Then ask your assistant:

"How visible is stripe.com in AI search?" "Run a Collimer scan on example.com and tell me the biggest AI-visibility gap." "Does ChatGPT cite acme.com when asked about its category?"

Your assistant will call collimer_scan and return the teaser.

The tool

collimer_scan(domain, email?)

Returns the teaser — score, confidence interval, top gap, report URL, and an unlock link. The full ranked fix plan and verification re-scan are never returned; they unlock with a free account.

Configuration (env)

What is Collimer?

Collimer measures and improves how often AI answer engines cite your brand — generative engine optimization (GEO), the AI-search successor to SEO. This MCP server is the free front door: scan any site from inside your assistant, see the score and the biggest gap, then open the full report and get your ranked fix plan at collimer.com. API docs: app.collimer.com/docs.

Privacy

The server calls Collimer's public scan API for the domain you pass and returns the result. It doesn't read your files or your conversation. Scans carry a source tag (default mcp) so we can see which channels are useful — no personal data beyond the optional email you provide.

Develop

npm install
npm run build          # → dist/
npm run smoke          # real scan against prod, prints the teaser
npm run test:protocol  # spawns the server + calls the tool over MCP/stdio

License

MIT © Sandcastle Labs