Log inGet Started

Your AI visibility, inside your coding agent

GetCited.me runs an MCP server over Streamable HTTP. Point Claude Code or Cursor at it and your agent can read how AI assistants answer questions about your brand, which of your pages get cited and what is missing — then write the exact files that fix it and check they went live. Every tool is a read, scoped to what you grant, and none of them spends a credit.

Updated

Scan your siteSee pricing

Why this one

Data your agent cannot get anywhere else

Most SEO MCP servers proxy your own Google account back to you — Search Console, Analytics, numbers you already own and any developer could wire up in an afternoon. This one serves what only a tracker of AI answers has: what the assistants actually said about you this morning, and what to do about it.

  • Answer-level visibility across Gemini, ChatGPT, Perplexity, Claude and Google AI Overviews — mention rate, the move since yesterday, and which brands are being named instead of you.
  • Which sources and which of your pages those answers cite, where you are absent from the answer entirely, and the coverage gaps behind it.
  • Deployable files with the repository path each one belongs at — and a live re-probe that says whether your deploy actually landed.

What the server is

  1. Nothing to install

    The server runs on our side at a fixed HTTPS address. Your client needs two things: that URL and an API token in an Authorization header. No package, no local process, no build step, and no session state to keep alive.

  2. Scoped to what you grant

    The agent authenticates with an API token you create in Settings, carrying only the scopes you tick and, if you want, pinned to a single project. A tool the token has no scope for refuses the call and names the scope it needed, so an under-granted token fails loudly instead of behaving oddly.

  3. Nothing here spends

    All eight tools are reads, or a live re-probe of your own site. None of them calls a model, burns an answer credit or touches a monthly quota, so an agent left running in a loop cannot run up a bill.

  4. The same data as the screens

    The tools wrap the code the product itself uses — the same audit artifacts, the same draft export, the same deployment check — rather than a parallel implementation. Your agent and the dashboard you are looking at cannot disagree.

The eight tools

Every call checks the token's scopes first, so an agent only reaches what you granted it.

ToolWhat your agent getsScope
list_websitesWhich projects this token can see, with their host, brand, plan and billing status.websites_read
get_visibility_snapshotGEO score and its breakdown, today's and yesterday's mention rate, the day-over-day move in percentage points, and the brand leaderboard for the answers we tracked.visibility_read
list_coverage_gapsQuestions where you are missing from the answer, or named but cited as the wrong site, each tagged with its remedy: write content, or clarify the entity.visibility_read
get_citation_targetsThe sources that answers in your category already cite — directory listings to claim, videos worth pitching, authors worth contacting.visibility_read
get_page_inventoryEvery page we know about with its citability score and band, the score it could reach, its quick-win points and the specific fixes.websites_read
get_fix_packAudit artifacts and Entity Clarify output as files — llms.txt, ai.txt, robots.txt, sitemap, JSON-LD, FAQ markup — each with the path it belongs at and suggested repository folders.audits_read
get_article_markdownYour article drafts, and the full article as markdown when you name one.content_read
verify_deploymentRe-fetches llms.txt, ai.txt and robots.txt from your live site and diffs them against what the audit expected.audits_write

Connect it

  1. Create an API token

    Settings, MCP server tab, create a token. Tick the scopes the agent needs and, if you want, pin it to one project. The value is shown exactly once — copy it then; only a hash is stored.

  2. Paste the config

    Both clients read a JSON file. Drop the matching block below into your repository, or into your user config to have the server in every project, and replace the placeholder with your token.

  3. Restart the client

    Claude Code and Cursor read MCP configuration at start. After a restart the server shows up with its eight tools; ask your agent to list them if you want to see it for yourself.

  4. Ask it something

    Start with a visibility snapshot, then work down the prompts below. Your agent already knows your repository — the missing half was knowing what AI assistants say about it.

Server URL

https://api.getcited.me/mcp

Claude Code

Save as .mcp.json in your repository root, or add the block to ~/.claude.json to have it in every project.

{
  "mcpServers": {
    "getcitedme": {
      "type": "http",
      "url": "https://api.getcited.me/mcp",
      "headers": {
        "Authorization": "Bearer citk_YOUR_TOKEN"
      }
    }
  }
}

Claude Code

Or add it from the terminal:

claude mcp add --transport http getcitedme https://api.getcited.me/mcp --header "Authorization: Bearer citk_YOUR_TOKEN"

Cursor

Save as .cursor/mcp.json in your repository, or ~/.cursor/mcp.json to have it in every project.

{
  "mcpServers": {
    "getcitedme": {
      "url": "https://api.getcited.me/mcp",
      "headers": {
        "Authorization": "Bearer citk_YOUR_TOKEN"
      }
    }
  }
}

Replace citk_YOUR_TOKEN with a token you create in Settings — the value is on screen only once, right after it is created.

Verified with Claude Code and Cursor: both send the Authorization header this server expects. claude.ai on the web connects through OAuth, which this server deliberately does not offer, so it will not work there. The Claude desktop app is not something we have tested yet — steps for it will appear here once it is verified, and not before.

Worth saving

Prompts to start with

Each one is a whole task, not a tool call. Replace the domain with your own project and keep the ones that earn their place in your week.

  • The Monday visibility check

    Answers whether you gained or lost ground in AI answers since yesterday, and who took the space.

    Using the getcitedme MCP server: list my websites, then get the visibility snapshot for example.com. Report the GEO score with its breakdown, today's mention rate against yesterday's, and the brands above me on the leaderboard. If the day-over-day move is negative, say which part of the breakdown fell and what changed.

    Weekly · Mondays · read-only

  • Triage the coverage gaps

    Separates questions that need a page written from questions where the model already knows you but cites the wrong site.

    Using the getcitedme MCP server, list the coverage gaps for example.com. Group them by remedy: the ones tagged content and the ones tagged entity_clarify. For the top three content gaps, tell me what a page would have to answer to close them, and check this repository for a page that already half answers it before proposing a new one.

    Weekly · read-only

  • Write the fix pack into the repo

    Turns the audit into actual files at actual paths, instead of a checklist you re-read every month.

    Using the getcitedme MCP server, get the fix pack for example.com. For every artifact marked deployable, write it to the targetPath it names, using suggestedRepoPaths to pick the right folder in this repository. Skip anything marked reference_only — that is what is already live. Show me the full diff before you commit anything.

    After each audit

  • Prove the deploy landed

    Catches the classic failure: the file is in the repository, the build ships, and the live site still serves the old one.

    My deploy for example.com is live. Using the getcitedme MCP server, run verify_deployment and tell me which of llms.txt, ai.txt and robots.txt are actually being served, and how each differs from what the audit expected. If anything is missing or stale, find out which part of this repository's build is dropping it.

    After every deploy

  • Spend an hour where it pays

    Ranks your existing pages by how much citability a small edit would buy, rather than by traffic.

    Using the getcitedme MCP server, get the page inventory for example.com and sort it by quickWinPoints. Take the top five pages, map each one to its source file in this repository, and apply the topFixes listed for it. Tell me the score each page is at, the score it could reach, and what you changed.

    Monthly · before a content sprint

  • Get cited where the answers already look

    Turns citation targets into a concrete outreach list instead of a generic backlink chase.

    Using the getcitedme MCP server, get the citation targets for example.com. Give me the directory listings I am missing, the videos worth pitching and the authors worth contacting, ordered by how often each source is already cited in answers about my category. For each one, draft the two sentences about us that source would need.

    Monthly

  • Ship the draft you already have

    Moves a finished article out of the product and into your repository in the format your site actually publishes.

    Using the getcitedme MCP server, list the article drafts for example.com and fetch the markdown for the one I name. Convert it into a post for this repository's content directory with the front matter this site uses, fix the internal links to real paths that exist here, and open a pull request. Do not invent facts or statistics that are not in the draft.

    When a draft is ready

Frequently Asked Questions

No. The server runs on our side over Streamable HTTP, so your client needs only the URL and an API token in a header. There is no package to add, no local process to keep running, and because the transport is stateless there is no session to lose.
No. All eight tools are reads or a live re-probe of your own site — none of them calls a model, burns an answer credit or touches a monthly quota. Anything that spends money, such as running an audit or generating an article, is deliberately not exposed here and stays behind an approval you give in the product.
Claude Code and Cursor are verified and documented above. claude.ai on the web connects to remote servers through OAuth, which this server does not offer, so it will not work there. The Claude desktop app we have not tested yet, so we publish no steps for it rather than steps that might not work.
A request with no valid token is refused with 401 and the client reports the server as failing to connect. A valid token that lacks the scope a tool asks for gets a clear refusal naming the missing scope, and a token pinned to specific projects is refused on any other project rather than silently returning nothing.
Request the discovery route on the server URL with /info on the end. It is public — no token needed — and returns the protocol version, the transport and the current tool names, which is also the fastest way to confirm the address you pasted is right.

See how AI answers about your market

Free scan: market map, AI visibility snapshot, and a GEO score in about a minute. No card, no call.

Scan your site free