Cite42: AI visibility and SEO data MCP server

Cite42 is an MCP server for AI visibility, also called generative engine optimization (GEO), and SEO data. It works in Claude Code, Codex, Cursor, Claude Desktop and ChatGPT. Your agent can ask the AI models behind ChatGPT, Claude, Perplexity and Gemini, plus Google AI Overviews, a buyer question and see whether the answers mention your brand and which sources they cite. It can also capture the answer from the ChatGPT app itself, measure from a chosen country, check where a site ranks in Google, and pull keyword search volume, Google Trends, Reddit and YouTube data.

You pay per call from a prepaid USD balance. There is no subscription, and new accounts get $1 of free credit. See pricing.

What you can do with it

Check whether AI answers mention your brand

Ask a buyer question such as "best invoicing software for freelancers" and see which brands each AI model names, in what order, and how often (cite42_rankings). Compare your brand with named competitors (cite42_compare), or see how the answers describe it (cite42_sentiment).

Measure the ChatGPT app itself, from any country

The default chatgpt surface is the OpenAI API with web search. Add chatgpt_app to models and Cite42 also asks the question in the ChatGPT app itself and returns what that interface answered. It is opt-in and never part of the defaults or a full sweep, because one capture can take up to two minutes. A single tool call waits 30 seconds for it and then returns the OpenAI API answer in its place, marked fallback on the result; trackers wait the full time. The app costs $0.05 per prompt, or $0.07 for sentiment.

Pass country (for example "gb") to any AI-model tool to measure from that market. OpenAI, Claude and Perplexity search with it as an approximate location, and Google AI Overviews and the ChatGPT app are read from that country, in its main language or the language you pass. Gemini's API has no country setting, so it runs without one. Each surface's measurement entry says what it actually used.

See which sources AI answers cite

cite42_citations lists the URLs the AI models cited for a question and can check whether your page or domain is among them. For the full text, cite42_search returns each model's answer with its sources. cite42_recommendations then compares your page with up to two of the cited pages.

Keyword search volume, CPC and competition

cite42_keywords returns monthly search volume, CPC and competition for a keyword, with optional related keyword ideas. cite42_keywords_batch checks up to 50 keywords in one call.

Where a site ranks in Google

cite42_keyword_rankings checks Google's top 100 organic results for one to five keywords in one of 40 countries, on desktop or mobile. For each keyword it returns your site's position and ranking page, up to five competitors' positions, the top 10, and whether the AI Overview cites your site. It costs $0.09 per keyword checked, and a keyword that fails is free.

Google Trends

cite42_trends returns Google Trends interest over time, related and rising queries, and a trend label.

Reddit and YouTube research

cite42_reddit_trends finds relevant Reddit discussions with cited summaries. cite42_youtube_trends finds rising videos, creator angles and title patterns.

Track prompts weekly or monthly

A tracker reruns up to 30 prompts every week or month and keeps the history, so you can see when AI answers start or stop naming your brand. A tracker can include the ChatGPT app and measure every prompt from one country. A keyword rankings tracker checks up to 100 keywords for one site on the same schedule.

Run research workflows in one call

Six workflows combine the tools above: content opportunities, topic demand, AI prompt maps, competitor content, AI citation gaps and content briefs.

Requirements

• A Cite42 account, free to create at cite42.dev
• For the API-key setup only: Node.js 22.13 or newer and a Cite42 API key

Installation

Most clients connect to the hosted server at https://mcp.cite42.dev and sign in with your Cite42 account in the browser. There is no API key to create or paste, and nothing to install locally. For CI, servers, or machines without a browser, run this package with an API key instead; see API key setup.

Claude Code

claude mcp add --transport http --scope user cite42 https://mcp.cite42.dev && claude mcp login cite42

Your browser opens to sign in. Then start Claude Code, or restart it if it was already open, and check with /mcp.

Or install Cite42 as a plugin from inside Claude Code:

/plugin install cite42 --marketplace Cite42/mcp

Claude Code versions before 2.1.275 need two steps: /plugin marketplace add Cite42/mcp, then /plugin install cite42@cite42. Then open /mcp, pick the Cite42 server and sign in.

Codex CLI

codex mcp add cite42 --url https://mcp.cite42.dev

Codex opens your browser to sign in. If it doesn't, run codex mcp login cite42. Then start a new session.

Cursor

Add to Cursor, or add this under Settings → MCP, or in ~/.cursor/mcp.json:

{
  "mcpServers": {
    "cite42": { "url": "https://mcp.cite42.dev" }
  }
}

Sign in with your Cite42 account when Cursor asks.

Claude Desktop and Claude.ai

Add to Claude opens the connector dialog with Cite42 filled in. Or open Settings → Connectors → Add custom connector, name it Cite42, and paste https://mcp.cite42.dev. Click Connect and sign in. The connector appears in both Claude Desktop and Claude.ai.

ChatGPT and Codex Desktop

Turn on Developer mode in Settings → Security and login, then add a plugin under Settings → Plugins with the MCP server URL https://mcp.cite42.dev and OAuth.

API key setup (CI, servers, no browser)

Create a key at www.cite42.dev/app/keys (it looks like cite42_live_... and is shown once), then replace cite42_live_your_key_here below. The client starts this package on demand; you don't need to keep it running.

# Claude Code
claude mcp add --scope user cite42 -e CITE42_API_KEY=cite42_live_your_key_here -- npx -y @cite42/mcp

# Codex CLI
codex mcp add cite42 --env CITE42_API_KEY=cite42_live_your_key_here -- npx -y @cite42/mcp

For Cursor, save this to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "cite42": {
      "command": "npx",
      "args": ["-y", "@cite42/mcp"],
      "env": {
        "CITE42_API_KEY": "cite42_live_your_key_here"
      }
    }
  }
}

Restart the client afterwards: a new terminal or claude/codex session, or fully quit and reopen Cursor.

Verify it's running

• Claude Code: run claude mcp list and check that cite42 shows as connected, or type /mcp inside a session.
• Codex CLI: run codex mcp list.
• Cursor: look for cite42 in the MCP/tools indicator.
• Claude Desktop / Claude.ai: open Settings → Connectors and confirm Cite42 is connected.

If it doesn't appear, see Troubleshooting.

Try it: example prompts

After connecting or restarting, just ask in plain language. Your agent picks the right Cite42 tool automatically.

• "Use Cite42 to check which brands AI models recommend for 'best project management software'."
• "Which sources do AI models cite when asked about CRM tools for startups? Is mycompany.com among them?"
• "Check which brands the ChatGPT app itself names for 'best CRM for startups', measured from Germany."
• "Get search volume, CPC, and related keywords for 'ai seo tools'."
• "Where does mycompany.com rank in Google for 'ai seo tools' and 'geo tools' in the UK, on mobile?"
• "Is interest in 'ai seo' rising on Google Trends? Show the related rising queries."
• "What are people on Reddit complaining about regarding email marketing tools?"
• "Run a content brief for the topic 'ai website builder'."
• "Map the prompts worth tracking for our category, then create a weekly Cite42 rankings tracker for the best five. Show me the estimated cost in USD per run before activating it."

Example: what a session looks like

You ask:

Use Cite42 to find AI-cited competitors for "best invoicing software for freelancers".

Your agent calls cite42_citations and answers something like:

Across ChatGPT, Perplexity, and Gemini, the most-cited sources are:

1. freshbooks.com (cited by 3/3 models)
2. wave.com (cited by 2/3 models)
3. zoho.com/invoice (cited by 2/3 models)
4. bonsai.com (cited by 1/3 models)

Your domain wasn't cited by any model. FreshBooks is cited mainly via comparison/"best of" listicles. A citation-gap analysis (cite42_find_ai_citation_gaps) can show which queries you're missing from.

That is the three default surfaces; ask for fullSweep to add Claude and Google AI Overviews, or name chatgpt_app to add the ChatGPT app itself. The exact brands and sources change by query and model. A good result should show the tool used, model/source evidence, and a short summary you can act on.

FAQ

Is there an MCP server for AI visibility?

Yes. AI visibility is how often AI assistants name or cite your brand when people ask about your category, and Cite42 is an MCP server that measures it. Connect it to Claude Code, Codex, Cursor, Claude Desktop, Claude.ai or ChatGPT, and your agent can ask the AI models behind ChatGPT, Claude, Perplexity and Gemini, plus Google AI Overviews, the questions your buyers ask and see which brands and sources come back.

How can I track my brand's visibility in ChatGPT and other AI search engines?

Connect Cite42 and ask something like "Use Cite42 to check which brands AI models recommend for 'best CRM for startups'." Your agent calls cite42_rankings, which reports which brands each answer names, in what order and how often. It checks ChatGPT, Perplexity and Gemini by default; ask for a full sweep to add Claude and Google AI Overviews. To follow it over time, a tracker reruns up to 30 prompts every week or month and keeps the history.

Does Cite42 read the real ChatGPT app or the OpenAI API?

Both. The default chatgpt surface asks the OpenAI API with web search enabled, which is fast and cheap. The opt-in chatgpt_app surface asks the question in the ChatGPT app itself and returns what that interface answered, so you can compare the two. It costs $0.05 per prompt and is never part of the defaults or a full sweep. A single call waits 30 seconds for the app and then returns the OpenAI API answer in its place, labelled as a fallback; trackers wait the full capture time.

Can I measure AI answers from a specific country?

Yes. Pass country with an ISO-2 code such as "gb" or "de" to any AI-model tool or tracker. OpenAI, Claude and Perplexity search from that location, and Google AI Overviews and the ChatGPT app are read from Google in that market. Gemini has no country setting and runs unlocated. The measurement block on each result says which location was actually used.

How can I see which websites ChatGPT, Claude and Perplexity cite?

cite42_citations lists the URLs the AI answers cited for a question and checks whether your page or domain is among them. cite42_find_ai_citation_gaps finds the questions where AI answers cite your competitors but not you.

How do I get my brand mentioned by ChatGPT?

Cite42 doesn't change what AI models say, but it shows you where to work. AI answers draw on the pages they cite, so start by finding those pages with cite42_citations. cite42_recommendations then compares your page with up to two of the cited pages and suggests changes, and a tracker shows whether your brand starts to appear.

Is there an MCP server for keyword research with real search volume?

Yes. Ask your agent for a keyword's search volume and it calls cite42_keywords, which returns monthly search volume, CPC and competition, with optional related keyword ideas. cite42_keywords_batch checks up to 50 keywords in one call.

Is there an MCP server for Google rank tracking?

Yes. Ask "Where does example.com rank for 'ai seo tools' in the UK?" and your agent calls cite42_keyword_rankings, which checks Google's top 100 for up to five keywords in one of 40 countries and returns the position, the ranking page, competitor positions, the top 10 and whether the AI Overview cites the site. It costs $0.09 per keyword. A keyword_rankings tracker rechecks up to 100 keywords weekly or monthly.

Is there a Google Trends MCP server?

Yes. cite42_trends returns Google Trends interest over time, related and rising queries, and a trend label, so you can ask "Is interest in 'ai seo' rising?" and get an answer from the data. cite42_reddit_trends and cite42_youtube_trends cover Reddit discussions and rising YouTube videos.

How do I find content gaps between AI answers and keyword demand?

cite42_find_content_opportunities combines AI answers, keyword demand, Google Trends, Reddit, YouTube and citations in one call to find topics you haven't covered.

Can I check AI visibility without a monthly subscription?

Yes. Cite42 charges per call from a prepaid USD balance, and there is no subscription. A rankings or citations check costs $0.13 on the three default surfaces (ChatGPT $0.05, Gemini $0.05, Perplexity $0.03) and $0.20 with all five. Failed calls are not charged, and new accounts get $1 of free credit. See cite42.dev/pricing, or ask your agent to run cite42_pricing.

Is there an AI visibility API for developers?

Yes. The same tools are available as a REST API at the same per-call prices, so you can build AI visibility checks into your own app. See the REST API docs.

Are the answers live?

Each answer comes from a real call to that AI model. Cite42 caches answers for 24 hours, shared across customers, so an identical prompt on the same surface can return an answer from up to a day earlier. Scheduled tracker runs are a week or a month apart, so they always fall outside that window.

Do I need an API key?

Not for the hosted server at https://mcp.cite42.dev, where you sign in with your Cite42 account in the browser. You only need a key to run this npm package, for example in CI or on a server without a browser.

Is it open source?

This MCP client is MIT licensed. The data comes from the hosted Cite42 API, which is not open source.

Tools

Cite42 exposes 30 MCP tools: 4 free account tools (including the compatibility alias), 12 data/report tools, 6 research workflows, and 8 tracker tools.

Account & pricing (free)

• cite42_balance: check the connected account’s remaining Cite42 balance in USD.
• cite42_credits: alias of cite42_balance for older integrations.
• cite42_pricing: get current per-tool prices and estimate a planned set of calls.
• cite42_usage: read recent call status and cost together with the remaining balance.

AI search visibility

• cite42_search: run a query against ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews, plus the ChatGPT app itself (chatgpt_app) when you ask for it, optionally from one country. Returns each model's answer plus cited sources.
• cite42_citations: aggregate which URLs the AI models cite for a query. Can also check whether a specific URL is cited.
• cite42_rankings: measure how brands rank in AI answers, including mention rate, average position, and per-model breakdown.
• cite42_compare: compare one brand against competitors in AI answers.
• cite42_sentiment: score positive, neutral, and negative sentiment for a brand, with supporting phrases.
• cite42_recommendations: page-level recommendations. A page URL alone gets a free technical check with suggested buyer questions; with a buyer question it measures citations and compares your page with up to two pages the AI surfaces cited, charged only when it includes verified page-comparison advice (plus the citations check it ran).

SEO & trend data

• cite42_keywords: get seed search volume, CPC, and competition, with optional related keyword ideas.
• cite42_keywords_batch: check volume, CPC, and competition for two to 50 seed keywords in one request; billed by outcome (call cite42_pricing for rates).
• cite42_keyword_rankings: where a site ranks in Google's top 100 for one to five keywords in one country, desktop or mobile: position, ranking URL, competitor positions, the top 10 and whether the AI Overview cites the site. $0.09 per keyword checked; a keyword that fails is free.
• cite42_trends: get Google Trends interest over time, related and rising queries, and a trend label.
• cite42_reddit_trends: find relevant Reddit discussions with cited summaries and read up to three pages through Gemini URL context. Returns page titles, authors, post excerpts, relative ages, available scores/comment counts and sampled human comments. Pages may be cached and model extraction is not independently verified. Exact dates, growth and quote-based insight arrays remain unavailable. Read relevantThreads; risingThreads is a deprecated compatibility alias. Check coverage and warnings for missing fields and out-of-window threads.
• cite42_youtube_trends: find YouTube rising videos, creator angles, title patterns, and opportunities.

Research workflows (multi-source, one call)

• cite42_find_content_opportunities: find content gaps across AI search, keywords, trends, Reddit, YouTube, and citations.
• cite42_analyze_topic_demand: combine keyword data, trends, AI answers, and audience conversations for a topic.
• cite42_map_ai_prompts: discover and cluster buyer prompts worth tracking in AI search. Its prompts are what cite42_tracker_create takes, so the two chain naturally.
• cite42_analyze_competitor_content: analyze AI rankings, comparisons, citations, and topic coverage for competitors.
• cite42_find_ai_citation_gaps: find queries where AI answers cite competitors but not your brand, domain, or URL.
• cite42_generate_content_brief: collect AI answers, citations, keyword demand, and social signals into brief-ready data.

Trackers

• cite42_tracker_create: create a draft weekly or monthly tracker — one measurement applied to a list of prompts, with the AI surfaces (including chatgpt_app) set once for all of them. No prompt list yet? Run cite42_map_ai_prompts first. Creating a draft is free and does not schedule calls. Set primitive to keyword_rankings with up to 100 keywords for one domain to track Google positions instead, and country to measure an AI tracker from one market.
• cite42_tracker_update: replace the whole tracker definition — name, cadence, measurement, prompts and configuration. A tracker read returns the same shape, so read it, change what you need, and send it back. Updating is free.
• cite42_tracker_activate: activate or resume recurring runs. You must pass the exact estimatedCostMicroPerRun from the latest tracker record as maxCostMicroPerRun; activation authorizes billed scheduled runs until the tracker is paused. The response carries the balance, the cost per run and runsDryAt — the date the balance stops covering your active trackers — and activation is refused when the balance cannot cover even the next run. When tracker emails are on, activation also sends a setup summary.
• cite42_tracker_pause: stop future scheduled runs while preserving configuration and history. Pausing is free.
• cite42_tracker_delete: permanently delete a draft or paused tracker and its run history. Active trackers must be paused first, and deletion is rejected while a run is queued or running. Financial usage records remain available.
• cite42_tracker_run: queue one manual run of every saved prompt. Each prompt is billed as its own tool call at the normal rate, and completion sends the same current-results email as a scheduled run—even for a baseline or an unchanged result.
• cite42_trackers: list trackers, schedules, statuses, cost estimates, and last/next run times. Reading the list is free.
• cite42_tracker_report: read bounded stored run history with results, changes, status, sampled time, and costs (latest 8 by default, up to 20). Reading stored reports is free.

Configuration

• CITE42_API_KEY: required for the local package. Your API key from www.cite42.dev/app/keys.
• CITE42_API_BASE: optional. Defaults to https://www.cite42.dev/api/v1.

You can also pass these as CLI flags instead of environment variables:

npx -y @cite42/mcp --api-key=cite42_live_your_key_here --api-base=https://www.cite42.dev/api/v1

Billing

Tool and workflow calls charge your Cite42 balance in USD. Tracker management and stored-history reads are free; every scheduled or manual tracker run bills one call per saved prompt at the normal rate. Activating or resuming a tracker requires explicit confirmation of the current maximum cost in USD per run and authorizes recurring billed calls until you pause it. Calls fail with a clear error if the API key is missing or your balance runs out. See www.cite42.dev/pricing.

Troubleshooting

• Server not listed after setup: make sure you fully restarted the client (new terminal for CLI clients, full quit for desktop apps).
• "No Cite42 account is linked to this sign-in": you signed in with a different account than the one you use at cite42.dev. In Claude Code run claude mcp logout cite42, then claude mcp login cite42 and pick the same sign-in method as on the dashboard.
• command not found: npx: install Node.js 22.13+ from nodejs.org, then restart your terminal.
• Tool calls fail with an auth error: check the key was passed correctly (CITE42_API_KEY, no quotes or trailing spaces) and is active at www.cite42.dev/app/keys.
• Tool calls fail with an insufficient-balance error: add funds at www.cite42.dev/app.

Development

This repository is the source of the @cite42/mcp package on npm. It is a thin client: each tool calls the hosted Cite42 API, which runs the AI and data queries.

npm ci
npm run build
npm test            # unit tests
npm run test:stdio  # runs the built executable against a local fake API; no key, no paid calls

Report bugs in GitHub issues or email [email protected].

Links

• Website: Cite42: AI visibility and SEO data for AI agents
• API keys: Create a Cite42 API key
• Docs: Cite42 API & MCP documentation
• Pricing: Cite42 pay-per-call pricing
• npm: @cite42/mcp

USD billing

Every paid result carries a cite42Billing block with the amount charged and the remaining Cite42 balance. Its chargedUsd and remainingUsd values are decimal strings in USD. Credits are retired: since 0.8.0 the package no longer returns cite42Credits, and cite42_credits answers exactly like cite42_balance.

cite42_pricing always returns the USD price table; an older currency argument is ignored. REST clients can use GET /v1/pricing?currency=USD. Tracker activation still accepts the exact integer maxCostMicroPerRun returned by the API; present its USD equivalent to users.

Optional AI visibility recommendations

cite42_rankings, cite42_compare, cite42_citations and cite42_sentiment include a metrics block. Show mention/citation percentages with their counts (for example, 66.7% — 2 of 3 checked answers). Sentiment percentages describe AI-estimated tone, not the percentage of positive answers. Failed, empty and incomplete observations are excluded; no usable sample is unavailable, not zero. Only comparable tracker runs provide automatic history. Each reported run includes metrics grouped by primitive and brand/watched URL, with current, previous, previousRunId, and changes. Each group states models and methodologyVersion. Its comparison is comparable, baseline (no earlier run, or a new measurement baseline after an old-format run), or unavailable; comparisonReason explains skipped comparisons. A successful run with unavailable comparison data is not a failed run, and its email does not claim "no changes". Completed manual-run replays include the same immediate predecessor as tracker reports. Citation groups preserve the original watched URL; normalized identities are only used for matching. Counts sum distinct prompts: four prompts × five surfaces can supply 20 answers. Comparisons require the complete prompt/configuration set, methodology and complete coverage to match the preceding run. Individual prompts remain small samples (at most six surfaces); percentage-point changes are withheld below 10 answers per group in each run. This guardrail does not establish statistical significance.

cite42_recommendations is a separate, optional report. With no saved result, url alone runs a free technical check (robots.txt crawler access, indexing, rendering) and suggests buyer questions, and url with query first runs a citations check, priced like cite42_citations. Otherwise use one requestId returned by an earlier call, a user-owned trackerRunId, or a structured evidence array. Pass url for the page that should answer the prompt, or context.domain to look for a match in that site's sitemap, and query to choose a prompt when the evidence covers several. Optional context also accepts brand, audience and goal for this request only. The report fetches that page, its robots.txt and up to two pages the AI surfaces cited, as returned HTML without JavaScript. Check cite42_pricing for the current report price and obtain the user's agreement before running the add-on. It is charged only when the report includes page-comparison recommendations backed by quotes from the inspected pages; everything else is free.

For rankings evidence, provide context.brand to identify the tracked brand you want advice for. Without it, the measurements remain available but the tool asks which brand to focus on for free; it never guesses ownership or advises you to improve a competitor's offering. Comparison evidence already identifies its target through input.brand.

Sources must be complete and no more than 30 days old, within 20 observations and 64KB. Supported successful workflow steps can be reused. The tool does not inspect websites, retrieve host memory, perform fresh research, recommend keywords, or activate paid tracking. Suggestions are labelled inferred and do not establish website defects or causes. Tracker follow-up requires a cadence and timezone choice plus separate activation approval. Identical completed retries are free for two minutes; an in-flight duplicate is refused without another charge and can be retried shortly.