cyted
Docs · MCP

Use Cyted inside your AI assistant

Connect your Cyted audits to Claude, Claude Code, ChatGPT or Cursor with one URL. Ask how you compare with competitors, pull the fix list into your editor, mark fixes done and request a fresh audit.

Server URLhttps://getcyted.ai/api/mcp
At a glance

Connect once. Your assistant does the reading.

One tool only asks: request_audit_run sends you a link, and a fresh audit runs when you click Approve run.

  1. 1Connects with one URLSign in to Cyted once and click Allow
  2. 2Reads your auditYour score, who is ahead and the real answerslist_sitesget_auditget_answers
  3. 3Pulls your fix listThe full action plan, most severe firstget_fixes
  4. 4Marks fixes doneWrites a status to your fix list onlyset_fix_status

Cyted runs a remote MCP server, so the assistant you already use can read your audits for you. Add one URL to Claude, Claude Code, ChatGPT or Cursor, sign in to Cyted once, and ask about your AI visibility in plain words. In a code editor the assistant can also pull your fix list, change your site, and mark each fix done.

MCP access is part of the Pro plan. You need at least one completed audit before the tools have anything to return.

What can you ask it?

Anything your audit report answers. A few that work well:

  • "How is my site doing in AI answers, and who is ahead of me?"
  • "What should I fix first? Show me the high-severity fixes."
  • "What did ChatGPT actually say when it was asked for a web designer in Leeds?"
  • "Which pages do the engines cite instead of mine?"
  • "Fix the top site issue in this repo, then mark it done in Cyted."

The assistant picks the right tool itself. You never type tool names.

Connect Claude (web and desktop app)

  1. Open Customize, then Connectors.
  2. Click +, then Add custom connector.
  3. Name it Cyted and paste the server URL: https://getcyted.ai/api/mcp. Leave the advanced settings empty.
  4. Click Add, then Connect. A Cyted window opens: sign in and click Allow.
  5. In a chat, open the + menu, then Connectors, and switch Cyted on.

On a Team or Enterprise plan, an owner adds the connector under Organization settings, then Connectors, and each member clicks Connect.

Connect Claude Code

Run this in your terminal:

claude mcp add --transport http cyted https://getcyted.ai/api/mcp

Then start Claude Code, type /mcp, pick cyted and follow the sign-in in your browser. The /mcp panel shows Connected once it works. Claude Code refreshes the sign-in on its own after that.

Connect ChatGPT

ChatGPT connects to MCP servers through developer mode, on the web app.

  1. Open Settings, then Apps & Connectors, then Advanced settings, and switch on Developer mode.
  2. Back in Apps & Connectors, click Create.
  3. Name it Cyted, paste https://getcyted.ai/api/mcp and choose OAuth for authentication.
  4. Sign in to Cyted in the window that opens and click Allow.
  5. In a new chat, pick Cyted from the developer mode tools in the composer.

Connect Cursor

Add Cyted to ~/.cursor/mcp.json for every project, or to .cursor/mcp.json in one project:

{
  "mcpServers": {
    "cyted": {
      "url": "https://getcyted.ai/api/mcp"
    }
  }
}

Open Cursor's settings, find Cyted in the MCP list, and click the sign-in prompt next to it. Cursor opens Cyted in your browser: sign in and click Allow.

What happens when you connect?

Your assistant sends you to getcyted.ai. Sign in with your usual email or Google account and you see one screen, Connect Claude to Cyted (or whichever assistant you used), with an Allow button. Allowing it gives the assistant access to every site on your account. There is no password to copy and no API key to store.

If your account is not on Pro, that screen offers the upgrade instead of Allow. Access is checked on every call, so if Pro lapses the tools stop answering until it is back.

What can the assistant do, and what can it change?

There are eight tools: six only read, one writes a status to your fix list, and one asks you to approve a new audit without being able to start it.

Tool What it does Changes anything?
list_sites Your sites, each with its latest audit, its visibility score, the two rates behind the score, the change since the audit before, and its status No
get_audit One audit's results. The summary view gives your visibility score, the two rates behind it, per-engine rates and the questions asked. The competitors view gives the ranking of businesses that offer what you do, and how to beat the ones ahead. The site view gives site signals, page grade and crawlability No
get_score_history Every completed audit of a site, newest first: its score, the two rates behind it, what it asked about, how its answers were collected, and its change since the audit before it, or why the two are not compared. It also marks which audits compare with the newest one. Your dashboard charts every audit and marks with a dashed line each one that isn't compared with the audit before No
get_fixes The full action plan, most severe first, each fix with a severity and a status No
get_answers The real AI answers for one of the audit's questions: which engine, who was named, whether you were, and the answer text No
set_fix_status Marks one fix as todo, in progress or done Your fix list only
request_audit_run Creates an approval link for a fresh audit of a site Nothing until you approve
get_run_status Whether a requested run is pending, running, complete, failed, expired or cancelled No

Fix statuses set by the assistant are the same ones you see on the Actions tab of your report, so ticking a fix in either place shows in both.

What do the score numbers mean?

Every audit has one visibility score out of 100, and the assistant also gets the two rates it is made from:

  • Named without being asked for (unbranded): how often the engines named you when the question did not mention your business, such as "Who is the best plumber in Leeds?"
  • Named when asked for by name (branded): how often they named you when the question did, such as "Is Acme Plumbing any good?"

The score leans almost entirely on the first rate, because being named by an engine that was asked about you says little about being found. The weights come back with the audit as scoreWeights, and the methodology page explains them. The confidence interval the assistant sees is the one on the score itself.

Questions about your prices, and the Search Console queries we run for a connected site, are asked and their answers are kept, but they do not count towards the score, either rate, or the per-engine rates. get_audit marks every question with whether it counts.

An audit from before the score was split has one pooled rate and no parts. The assistant is told so (scoreBasis is pooled_legacy), and it gets no change figure between that audit and a newer one, because the two scores do not measure the same thing. For such an audit, and only for such an audit, the assistant also gets the counts behind the score (appearsIn answers that named you, out of sampleSize); for every other audit the counts come with the two rates instead.

The change figure (delta in list_sites) is also left out between two audits that asked about a different category or a different city, and between two audits judged by different rules for who counts as you and who counts as a competitor. When Cyted changes those rules, an audit from before the change and one from after it are not compared.

It is left out, too, between two audits whose answers were collected differently. Two audits compare only when each question was put to each AI the same number of times. Every audit asks each question three times, whether it is a weekly rerun or a run you start by hand, so the two compare. From 9 to 14 Nov 2026 weekly reruns asked once, so a weekly rerun from those days is not compared with a run that asked three times. Nor are two audits answered by different AI models or a different set of engines. Older audits did not record which AI models answered, so a site's first audit since we started recording them gets no change figure against the one before it. The dashboard's chart still shows every audit on one line: a dashed line marks each audit that isn't compared with the one before, and says why.

get_score_history gives the same verdict for each audit against the one before it, and also marks which audits compare with your newest one. Two marked audits can sit either side of one that is not marked, such as two weekly reruns from 9 to 14 Nov 2026, when weekly reruns asked once, with a run you started by hand between them. The assistant may compare those two weekly reruns, but not either of them with the run between.

What does the competitor ranking count?

The ranking in the competitors view counts businesses that offer what you do, in answers to questions that did not name you, plus one row for you (isSelf). AI assistants, review sites, media, funders and the like are left out. A name written as a web address counts as you only when it is your own address. If AI named you only when a question asked for you by name, you have no row in it. An audit from before this rule keeps the ranking it stored, which counts every name in every answer.

How do I use it to fix my site?

With Cyted connected in Claude Code or Cursor, open your website's repository and ask:

Get my Cyted fixes, fix the highest-severity one in this codebase, and mark it done when you have finished.

The assistant calls get_fixes, reads the evidence attached to the top fix, edits your pages, and calls set_fix_status. Review the change like any other pull request before it goes live. A fix marked done in Cyted means somebody changed the site. It does not prove the engines have noticed yet, which is what a fresh audit is for.

How do I run a fresh audit from the assistant?

Ask for one: "Run a new Cyted audit of my site." The assistant calls request_audit_run and gives you a link to getcyted.ai. Nothing runs until you open that link, signed in, and click Approve run. The page tells you how many audits you have left this month before you decide.

  • The link expires after 30 minutes. Ask again for a new one.
  • An approved run uses one of your 10 audits for the month. The allowance resets on the 1st.
  • Asking twice returns the same link, so the assistant cannot queue runs behind your back.
  • If a site already has an audit running, the assistant is told to wait.
  • If we do not know what the business does yet, the assistant is told so and gives you no link. Set it under Site settings on the site's dashboard, then ask again.

After you approve, the assistant can call get_run_status to see when the audit is complete, then read the new results.

What do the errors mean?

Every error comes back with a short note on what to do next, and assistants usually act on it without asking you. For reference:

Code Meaning What to do
not_pro The account is not on Pro Upgrade on the billing page, then try again
not_found The site, audit, fix or question does not exist on your account Ask the assistant to list your sites first
audit_running The audit is still in progress Wait a few minutes
run_in_progress A new run was requested while one is already going Wait for it to finish
needs_site_settings The site has no saved answer for what the business does, or a city that cannot be asked about, so a new run cannot start Set it under Site settings on the site's dashboard, then ask again
quota_exceeded All 10 audits this month are used Wait for the reset on the 1st
internal Something failed on our side Try again in a minute

How do I disconnect?

Remove the Cyted connector in your assistant: Connectors in Claude, claude mcp remove cyted in Claude Code, Apps & Connectors in ChatGPT, or delete the entry from mcp.json in Cursor. Your audits and fix list stay as they are on getcyted.ai.

Not on Pro yet? Run an audit first. Every tool on this page reads from it.