Connect with your Cited account.
No API key to create or manage. Add the Cited MCP URL to Claude Desktop, sign in, and the brands your account can reach are available through the connection. On Pro and above.
https://api.getcited.in/mcp-oauth
Your Cited data, inside the assistant you already use.
MCP is an open standard for connecting AI assistants to outside tools and data. Cited exposes your AI search data through it, so a supported assistant can answer questions about your brands, topics, sources and metrics without opening the dashboard. The data comes from the same store the dashboard reads.
Why put Cited inside your assistant?
Your AI search data should be where decisions get made. MCP lets your team ask in plain language, work across every brand at once, and go from a number straight to the evidence under it, without leaving the assistant they already have open.
- Ask, do not navigate
- “Which topics did we lose this month?” instead of four filters and a date range.
- Go from metric to evidence
- “Which sources are driving those losses?” The drill is a follow-up question, not another screen.
- Work across brands
- “Compare all the brands I manage.” An agency key covers its whole roster in one conversation.
Two ways to connect.
Which one you use depends on the assistant.
Connect with an API key.
For clients that use an authenticated MCP connection: Claude Code, Cursor, VS Code, Gemini CLI and ChatGPT. Create a key under Settings → API & MCP, then add the endpoint and the authorization header to your assistant. Key access is on Pro and above.
https://api.getcited.in/mcp Authorization: Bearer cited_sk_…
Then you just ask.
Ask in plain English. Your assistant chooses the Cited tools it needs and combines the results into an answer. Here is one question about topics and the calls it became.
“Across every brand I manage, which topics lost share this month, and which sources are driving those losses?”
list_brands()
→ 7 brands
cube_slice(brand_id, metric="sov", breakdown="topic", days=30)
→ share of voice per topic, per brand
cube_drill(brand_id, filters={"topic": "…"}, breakdown="domain")
→ the sources behind each topic that movedOne question. Several Cited tools. One answer. The assistant discovers your brands, slices the metric by topic, then drills into the sources behind the change, all from the same data layer the dashboard reads. There is no second calculation on the side, so the number your assistant gives you is the number on your screen.
Ten tools, read-only.
Ten read-only tools let an assistant discover your brands, query your metrics, drill into topics and the sources behind them, and read Cited's own published data. None of them can change anything in your Cited account.
whoamiShows which brands this key can read. Call it first, because every other tool needs the brand IDs it returns.Lookuplist_brandsLists every brand the key covers. One key can hold several, which is how an agency works through a whole client list in one conversation.Lookuplist_metricsLists the numbers you can ask for and the ways they can be broken down, so the assistant works from the real options instead of guessing.Lookuplist_dimensionsLists the values behind a filter, such as which platforms, topics or intents exist for your brand. The assistant builds a filter from what is really there.Lookupcube_sliceOne number over a chosen period, optionally broken down by one thing such as platform or topic. This is where the dashboard's own figures come from.Metricscube_drillThe same tool, one level deeper each time. Platform, then topic, then prompt, then the answer itself.Metricscube_category_mixWhere the sources behind the answers come from: your own site, competitors, earned media, aggregators, communities and everything else. Overall and per platform. A list of domains will not tell you this.Metricsget_cited_indexThe published Cited Index: Indian brands scored on how often five AI engines name them. Free and public.Dataget_index_source_mixWhat kinds of site the answers in the Index were built from.Dataget_ai_shelf_productsThe products AI answers actually name when someone asks what to buy.Data
Nine metrics, six ways to slice them.
Ask for the metrics you already use in the dashboard, then break them down by platform, domain, topic, intent, prompt or date. Cited works each one out in one place, so the assistant and the dashboard stay aligned.
- AI Visibility Score
visibility - How often answers mention you. Brand mentions divided by tracked answers, weighted across runs, on a 0 to 100 scale.
- Citation Rate
citation_rate - Of the answers that mention you, how many also cite your own site.
- Citation Share
citation_share - Your citations as a share of all citations in the same set.
- Citation Count
citation_count - How many citations to you are in that set.
- Share of Voice
sov - Your share of all mentions across you and the competitors you track.
- Avg Position
avg_position - Where in the answer you usually appear, averaged over the answers that name you. Lower is better.
- Net Sentiment
net_sentiment - Positive mentions minus negative mentions, divided by the sample, with the counts behind it.
- Mention Count
mention_count - How many times answers mention you in that set.
- Response Count
response_count - How many tracked answers are in that set.
Break any of them down by platform, domain, topic, intent, prompt or run_date. There is one exception. citation_rate only counts the answers that mention you, so it cannot be split by domain. Use citation_share when you want the view per source.
Questions about MCP
Full reference in the docs, or talk to us.
Do the numbers match the dashboard?
Yes. These tools read the same stored numbers the dashboard reads, through the same code. There is no second calculation that could drift away from it, which is why MCP waited until those numbers lived in one place.
Which assistants does it work with?
Any client that speaks MCP, and the connection method depends on the client. Claude Desktop needs no key at all: paste the sign-in URL and log in. Claude Code, Cursor, VS Code and Gemini CLI take the key as a header. ChatGPT's connector dialog has no field for a header, so the key goes in the URL instead, and it needs a paid ChatGPT plan. The MCP docs carry the current setup for each.
Is this an API I can build on?
It is an MCP server rather than a REST API. The ten tools are all of it, and there is nothing wider behind them for an MCP client to reach. Pulling metrics into your own systems over an API is a separate Enterprise arrangement. If your agent needs something the ten tools do not cover, tell us what you are trying to build.
Can one key cover several brands?
Yes, and that is the point for an agency. A key is granted a set of brands. Every call is limited to that set, so an assistant can work through the whole list in one conversation.
What can it change?
Nothing. The tools are read-only: they retrieve metrics, sources and published data, and they cannot modify your Cited configuration or the data underneath it. Anything that acts on your account runs through Cited's own execution layer, where it is metered and recorded.
What does it cost?
There is no separate MCP product and nothing to buy on top of your plan. MCP is on Pro and above, whichever way you connect: signing in from Claude Desktop, or creating an API key for the other assistants. The REST API is separate and stays on Enterprise.
Where does my data go?
To the assistant you connect it to. The server returns only data your Cited account already has, but that data does leave Cited for the assistant you chose. Treat the key like a password.
Put Cited inside your assistant.
Connect Cited to a supported MCP client, then ask about your brands, topics, sources and AI visibility without opening the dashboard.