Cited AI Visibility Check

What Cited AI Visibility Check MCP server does

Cited AI Visibility Check MCP server evaluates how three AI search experiences represent a business: ChatGPT through the OpenAI Responses API, Perplexity Sonar, and Gemini with Google Search grounding. It sends buyer-oriented questions to the engines and examines their responses for business mentions and website citations.

The main tool is check_ai_visibility. A business name is required. You can also provide a domain, up to 10 questions, a category, a location, and a list of engines. If questions are omitted, the category is used to create five general buyer questions. If no engine list is supplied, every engine with a configured API key is used.

Results identify whether the business appeared, its position when the answer contains a list, whether its domain or subdomains were cited, competitors mentioned in the response, source domains, and an estimated cost. The tool returns readable output and structured JSON for MCP clients that support structuredContent.

How it works

The Cited AI Visibility Check MCP server uses each provider's API rather than controlling the consumer web applications. OpenAI requests use the web_search tool, Perplexity includes search in Sonar, and Gemini uses Grounding with Google Search. Engines without keys are skipped and reported.

Business matching is case-insensitive and ignores punctuation. It treats ampersands and the word “and” as equivalent, removes common legal suffixes such as LLC, Inc, and Co, and also recognizes a mention of the supplied website domain. Website citation detection compares the supplied domain and its subdomains with URLs cited by the engine.

Competitors are inferred from list entries, headings, and bold text, so that part of the result is heuristic. Location is inferred from the question text; include the city or region in each question when local intent matters.

Setup and configuration

Install the server through Node.js 20 or newer. Add a stdio MCP entry that runs npx -y cited-mcp, then place one or more provider keys in its environment. The README provides configurations for Claude Desktop, Claude Code, and Cursor; other MCP clients can use the same stdio command and environment setup.

Supported provider variables are OPENAI_API_KEY, PERPLEXITY_API_KEY, and GEMINI_API_KEY. Optional model overrides are OPENAI_MODEL, PERPLEXITY_MODEL, and GEMINI_MODEL. The default models are gpt-5.4-mini, sonar, and gemini-3.5-flash-lite, respectively.

There is no Cited account or Cited usage fee. Each question produces one API call per selected engine, and the provider bills the configured account. Returned cost estimates use provider-reported token usage and available list pricing; Gemini grounding charges beyond its free allowance may not be included.

Limitations and notes

Cited AI Visibility Check MCP server reports a snapshot, not a stable ranking. Engine answers can change between identical runs, so trend analysis requires repeated checks over time. The API results may differ from the consumer ChatGPT, Perplexity, and Gemini applications because those products can use different models, personalization, and location signals.

Name matching is not semantic. Abbreviations or nicknames used by an engine will not match unless that form is supplied as the business name. Competitor extraction can miss names or include unrelated text. The response excerpt is included so an assistant can inspect the underlying answer.

A default run can query five generated questions across every configured engine. Costs depend on provider pricing, model choice, token usage, and search or grounding fees. Provider prices can change, so billing dashboards remain the authoritative source for charges.