DEV Community

Artyom Rabzonov
Artyom Rabzonov

Posted on • Originally published at automatelab.tech

We were about to pay $400/month for an AI citation dashboard, so we built one

#mcp #ai #automation #citation

Hosted AI citation dashboards (Profound, AthenaHQ, Otterly, Ahrefs Brand Radar) start at $295 to $499 per month. They tell you which URLs ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews, and Bing cite for any query.

That data is mostly available through each vendor's own API. The dashboard is the product; the data is commodity. So we built Citation Intelligence MCP - a free, self-hosted MCP server that does the same job, with 12 tools, no backend, and per-query costs measured in cents.

What it does

12 tools across three jobs:

Job Tools
Track citations check_citations, am_i_cited, ai_overview, cited_for
Predict citations predict_citation, compare_domains, wikipedia_mentions
Find leaks gsc_citation_gap, audit_sitemap, citation_trend, run_panel, track_queries

Three of the tools (predict_citation, cited_for, wikipedia_mentions) run on a local cache and cost zero. The rest use your own keys.

Install 

npx -y @automatelab/citation-intelligence
Enter fullscreen mode Exit fullscreen mode

Wire into Claude Code by adding to .mcp.json:

{
  "mcpServers": {
    "citation-intelligence": {
      "command": "npx",
      "args": ["-y", "@automatelab/citation-intelligence"]
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Bring your own keys for the paid tools:

export ANTHROPIC_API_KEY=...
export OPENAI_API_KEY=...
export GEMINI_API_KEY=...
export SERPAPI_API_KEY=...   # for Google AI Overviews
Enter fullscreen mode Exit fullscreen mode

Cost is roughly $0.01 to $0.03 per query at premium engines, so a daily 50-query panel runs you about $20/month vs $400 for the hosted equivalent.

The report that made us actually ship this

The one that got us is gsc_citation_gap. It joins Google Search Console data with AI citation status and returns pages where you rank in Google but get zero AI mentions. The gap is the editorial budget.

> gsc_citation_gap site:automatelab.tech --threshold-impressions 1000

URL                                          GSC impressions   Cited by
/glossary/claude-code/                                 4,238   0 engines
/glossary/llms-txt/                                    3,102   1 engine (Perplexity)
/glossary/json-ld/                                     2,847   0 engines
Enter fullscreen mode Exit fullscreen mode

Anything cited by zero engines and ranking on Google is a page that needs FAQ markup, query-shaped H2s, or a direct rewrite.

Why MCP-first

We did not want a dashboard. We wanted citation data inside the same agent loop that writes the post. An MCP server lives next to your editorial flow - the agent that drafts your next article can call predict_citation before writing the H2, then am_i_cited a week after publish.

The full launch write-up: https://automatelab.tech/launching-the-citation-intelligence-mcp/

Repo: https://github.com/automatelab/citation-intelligence

MIT licensed. No backend, no telemetry, no account.

Top comments (0)