What is Citation Intelligence MCP?

Citation Intelligence MCP is a free, self-hosted Model Context Protocol server that reports which URLs ChatGPT, Claude, Perplexity, Google AI Overviews, Gemini, and Bing cite for any query. It ships 12 tools through any MCP-compatible client (Claude Desktop, Claude Code, Cursor, Cline, Continue, n8n, LangGraph). Install with npx -y @automatelab/citation-intelligence.

How is Citation Intelligence MCP different from Profound, AthenaHQ, or Ahrefs Brand Radar?

The hosted AI citation dashboards (Profound, AthenaHQ, Otterly.AI, Ahrefs Brand Radar) start at $295 to $499 per month and target marketing teams that want charts and team seats. Citation Intelligence MCP is the opposite: a local server, no account, no centralized backend. The 12 tools are the surface; whatever client you point at the server is the UI.

How do I install Citation Intelligence MCP?

Install with npx -y @automatelab/citation-intelligence (requires Node 20 or later). Add the server to your MCP host config (claude_desktop_config.json for Claude Desktop, .mcp.json at your project root for Claude Code, ~/.cursor/mcp.json for Cursor) and set the API keys for the engines you want to use. Restart the host and the 12 tools appear.

Free · MIT licensed

Citation Intelligence MCP

Free MCP server that tells your agent what LLMs cite, across Perplexity, Claude, ChatGPT, Gemini, Google AI Overviews, and Bing.

npx -y @automatelab/citation-intelligence

Get it on GitHub npm package Read the launch post

What it does

Citation Intelligence MCP tells your agent who AI search cites for any query, across all six engines that currently ship citations: Perplexity, Claude, ChatGPT, Gemini, Google AI Overviews, and Bing. The same data hosted dashboards charge $99 to $499 per month for, exposed as 12 MCP tools your agent calls directly. Free, self-hosted, bring your own API keys.

Built for indie founders and content marketers who can't justify a Profound or AthenaHQ seat, SEO and GEO practitioners moving budget from rank tracking to AI visibility, agent builders who want citation data as a tool and not a dashboard, and engineering teams who want to know whether AI assistants cite their docs.

What you get out of it: track which of your URLs each AI engine cites across a saved query panel on whatever cadence you set, with citation rate and per-query deltas landing straight in the agent; surface queries where you rank top 10 in Google but are invisible in AI (the closest editorial wins, and no rank tracker shows them); score any URL for citation likelihood before you publish, with no LLM fired and no API spend; audit an entire sitemap worst-first so the next content rewrite hits the page that needs it most.

What it gives your agent

Twelve tools across citation lookup, predictive scoring, panel tracking, sitemap audit, and GSC integration. Most need at least one engine API key; predict_citation, cited_for, and wikipedia_mentions need zero keys.

Tool	What it does	Example prompt
`check_citations`	URLs cited by Perplexity, Claude, ChatGPT, Gemini, or Bing for a query.	"What does Perplexity cite for 'best AI citation tracker'?"
`am_i_cited`	Presence and rank for a domain across a query cluster. Returns citation rate and average rank.	"For these 20 queries, is example.com cited by ChatGPT?"
`ai_overview`	Google AI Overview presence and cited sources for a query, via SerpAPI.	"Is there an AI Overview for 'self-hosted GEO tool'?"
`cited_for`	Queries the domain has been cited for, from local cache. zero keys	"What has automatelab.tech been cited for in the last 30 days?"
`predict_citation`	Citation likelihood score (0-100) for a URL from public signals. No LLM fired. zero keys	"How likely is https://example.com/post to be cited by AI?"
`track_queries`	Save, load, or list named query panels (editorial watchlists).	"Save these 25 queries as a panel called 'editorial-watchlist'."
`run_panel`	Run a saved panel through `am_i_cited` and snapshot to disk.	"Run the 'editorial-watchlist' panel and snapshot today's results."
`citation_trend`	Time-series report of citation rate plus per-query gained or lost deltas.	"Show the citation trend for the 'editorial-watchlist' panel since April."
`compare_domains`	Side-by-side `predict_citation` across 2-10 URLs (you vs competitors).	"Compare my post against the two top-cited competitor posts."
`wikipedia_mentions`	Wikipedia articles that already link to a domain. zero keys	"Which Wikipedia pages link to example.com?"
`audit_sitemap`	Bulk `predict_citation` across every URL in a sitemap, worst-first.	"Audit https://example.com/sitemap.xml and show me the lowest 20 pages."
`gsc_citation_gap`	Join Google Search Console performance with AI citation status. Surfaces queries that rank in Google but are invisible in AI.	"Find queries where example.com ranks top 10 in Google but is not cited by Perplexity."

What makes it different

MCP-first, not dashboard-first

Hosted AI citation dashboards (Profound, AthenaHQ, Otterly.AI, Ahrefs Brand Radar) start at $295 to $499 per month and target marketing teams. This is the same data, exposed through MCP, free, with your agent as the UI.

Self-hosted, no backend

The server runs on your machine. No account, no proxy, no telemetry by default. API calls go from your process directly to the vendor and back.

Zero-key tools that still work

predict_citation scores citation likelihood from public signals (Wikipedia, schema.org, llms.txt, GitHub, content depth) without firing any LLM. Free, instant, repeatable.

Six engines, one tool surface

check_citations ships adapters for Perplexity, Claude, ChatGPT, Gemini, Bing, and Google AI Overviews. Set engine: auto and the first available key picks itself.

Local cache cuts repeat spend

Citation queries cache for 7 days, AI Overview queries for 1 day. Repeated lookups during a session, or across a weekly panel run, cost nothing after the first hit.

GSC join finds closest wins

gsc_citation_gap joins Google Search Console performance with AI citation status. Queries where you already rank in Google but are invisible in AI are the closest editorial wins.

Install in 3 steps

Run the server with npx (requires Node 20 or later) npx -y @automatelab/citation-intelligence

Add the server to your MCP host config (claude_desktop_config.json for Claude Desktop, .mcp.json at the project root for Claude Code, ~/.cursor/mcp.json for Cursor)

{
  "mcpServers": {
    "citation-intelligence": {
      "command": "npx",
      "args": ["-y", "@automatelab/citation-intelligence"],
      "env": {
        "PERPLEXITY_API_KEY": "pplx-...",
        "SERPAPI_KEY": "...",
        "ANTHROPIC_API_KEY": "sk-ant-...",
        "OPENAI_API_KEY": "sk-...",
        "GEMINI_API_KEY": "..."
      }
    }
  }
}

Set only the keys you have. Perplexity and Gemini have free tiers; SerpAPI gives 100 free Google AI Overview lookups per month. predict_citation, cited_for, and wikipedia_mentions need zero keys.

Restart your MCP host. The 12 citation-intelligence tools appear in the MCP panel.

Works with Claude Desktop, Claude Code, Cursor, Cline, Continue, n8n, LangGraph, and any MCP-compatible agent harness. Full reference: the GitHub readme.

FAQ

What does it cost?

The server is MIT-licensed and free. The only cost is the per-query spend at the engine vendors: roughly $0.01 to $0.03 per query at ChatGPT or Perplexity. predict_citation, cited_for, wikipedia_mentions, and the panel and trend tools cost nothing because they hit local cache or zero-key endpoints.

Does it store my API keys?

No. API keys are read from environment variables on the MCP process at startup. They are never logged, never persisted, and never sent anywhere except the vendor whose key it is. There is no central backend.

Which MCP clients does it work with?

Anything that supports stdio transport: Claude Desktop, Claude Code, Cursor, Continue, Cline, n8n (via the MCP node), LangGraph, custom agents built on the official MCP SDKs. The command and args pattern is the same across clients.

How is it different from Profound, AthenaHQ, or Ahrefs Brand Radar?

Those are hosted dashboards starting at $295 to $499 per month, built for marketing teams that want charts, alerts, and team seats. Citation Intelligence MCP is the opposite design point: a local server, no account, no centralized backend, no shared dashboard. The 12 tools are the surface; whatever client you point at the server is the UI. The trade is real, but if you already work inside an agent, this is the shape you want.

Where is the local cache stored?

~/.config/citation-intelligence/cache.json on macOS and Linux, the equivalent path on Windows. Default TTL is 7 days for citation queries and 1 day for AI Overview queries. Override the directory with CITATION_CONFIG_DIR. Delete the file any time to reset.

How do I track citation trends over time?

Save a query panel with track_queries, then run run_panel on a schedule (weekly is the typical cadence). Each run snapshots to disk. citation_trend returns the per-query deltas: which queries flipped from cited: false to cited: true since the first snapshot. That delta is the editorial-impact metric.

What does it not do?

It does not ship a hosted dashboard or charts. It does not pre-model hundreds of millions of prompts the way Ahrefs Brand Radar does. It does not track citations for academic papers (try citecheck). It does not promise a Wikipedia article will appear if you ask nicely; the Wikipedia signal is measured, but no actionable advice is emitted for it.

Want it wired into your editorial pipeline?

We use Citation Intelligence MCP daily on automatelab.tech: pre-publish citation gates, weekly panel runs, sitemap audits. If you want it set up and tuned for your stack, or a custom citation-tracking pipeline built end to end, we can do that.

Get in touch