AI visibility by API and MCP: Claude, ChatGPT or Looker Studio

A marketing engineer asks this when the numbers must leave the dashboard: into a Claude or ChatGPT session, a Looker Studio report, or a script. This guide states what an AI visibility API returns, what an MCP server adds, and what the Data Studio connector feeds. It shows three requests we made on our own project, with what each one returned. It ends with the tools that offer each surface. For the report those numbers go into, read our guide on reporting to clients. For the week around the reads, read our guide for in-house marketing teams.

What does an AI visibility API return?

An AI visibility API returns the stored answers of AI platforms, and the counts over them, to a program instead of a screen. A request names a window, and often a platform, a market or a prompt. The response carries four kinds of record:

1. Answers. The text one platform returned to one prompt, in one market, on one day.
2. Namings. The brands each answer names, each with a sentiment label and a recommendation label.
3. Citations. The pages each answer cites, with their domains.
4. Searches. The web searches a platform ran before it answered, where the platform reports them.

A rate is one count divided by another. Visibility is the answers that name the brand over all analyzed answers. An API that returns both counts lets the client compute the rate over any grouping it likes.

FullReach AI sells no REST API. It has two machine surfaces. The MCP server, on every plan, answers an AI assistant. The Data Studio feed, from the Advanced plan, answers a Looker Studio chart. Both read with an API key that an admin creates in the organization settings. Neither asks a platform anything live. Every number comes from answers already collected, so two reads a minute apart agree.

What does an MCP server add?

An MCP server lets an AI assistant call a tool's reads from a question in plain words, so nobody writes a request by hand. MCP stands for Model Context Protocol, the open standard that ChatGPT, Claude and other assistants use to call tools. The assistant reads the server's tool list, picks the tools that answer the question, and quotes the figures.

The FullReach AI server runs at https://mcp.fullreach.ai. On 30 October 2026, its manual listed 22 read tools, 10 write tools and 10 saved prompts. A read tool returns the figures of one dashboard page, with the same filters. A write tool changes prompts, suggested questions and competitors, and only for a key or a connection with write access. A saved prompt, such as a weekly pulse or an action plan, runs as a slash command in the assistant.

The manual gives the steps for 4 clients: ChatGPT, Claude, Claude Code and Cursor. A client connects in one of two ways:

1. Sign in and approve. ChatGPT, claude.ai and Claude Desktop add the server as a custom connector. The server opens a FullReach AI page, an admin signs in and approves, and the page names where the answers go.
2. An API key. Claude Code, Cursor and any client that sends a header pass the key as a bearer token.

A connection reads every project of the organization, or one project. Without write access, it cannot change anything. The AI Agent inside the dashboard answers from the same reads, beside every page, and links to the view that shows the figure.

AI AgentOverview · 7 days

Which competitor leads, and by how much?

Read Overview, Competitors

HubSpot. It appears in 57.8% of answers in the last 7 days. Pipedrive appears in 41.1%, 2nd of 5 brands.

Open Competitorslast 7 days · all platforms

What do three worked requests return?

We made three calls on our own project through the MCP server, each with a 28-day period. The project tracks the category we sell in, AI visibility tools, on 7 AI platforms. Each response named its window, 3 to 30 October 2026, so the table states it once.

Each response states its rates as fractions from 0 to 1, and the assistant turns them into percentages. Each response also names its denominator, so the assistant can state the base beside the rate. The reads cost no collection, because the server answers from stored answers. The get_answer tool returns one answer word for word, so the assistant can quote the sentence behind a number.

The same overview read found our own brand in 14 of the 8,953 answers. A brand owner who never runs such a read does not know that figure. A first look on your own brand states it before you connect anything.

How does the Looker Studio connector work?

Looker Studio is Google's free reporting tool, which Google called Data Studio, and the FullReach AI connector feeds it three tables of counts. The Brands table holds the answers that name each monitored brand, per day, platform and market, with sentiment and recommendation. The Prompts table holds the answers to each prompt, with its topic, persona and language. The Sources table holds the cited domains, with the brand's own site marked.

Every value is a count, so a chart can add rows. Visibility and Share of Voice are formulas the connector adds as fields, and Looker Studio computes them over any grouping. The feed applies a chart's filters on the server, so a brand filter never changes the answer count that every brand shares.

The connector runs on Google's servers. It reads with the key you paste, and it reads counts, never the text of an answer. One Google account holds one key for all its data sources, so one key serves the organization. A report to copy has four pages: Overview, Platforms, Prompts and Sources. The connector page links the setup guide and the report. Each of the three tables has a Market dimension, which returns the country of the prompt as a two-letter code, with ZZ for Worldwide. Our guide on markets shows how to roll several markets into one scorecard.

What limits apply to API and MCP reads?

Every plan includes the MCP server, and the Advanced plan adds the connector. The table states the caps as the manual listed them on 30 October 2026.

The manual's Plan limits page, read on 30 October 2026, states that every MCP response carries rate-limit headers. A client reads them to pace itself before the server refuses a call. The same page states that the connector retries the charts past the 3 at once on its own. A key belongs to the organization, not to a person, and works until an admin revokes it. The manual's page on connecting a client, read the same day, states that a connection belongs to the admin who approved it. The connection ends when that admin leaves the organization or stops being an admin.

Which AI visibility tools offer an API or MCP?

The table reads each plan's data access from the record this site keeps, with the day we read the tool's pages. FullReach AI is our own product, and it comes first.

Our Profound review states what its trial and its paid plan give for API access, and what its pages say about an MCP server.

What to do next

1. Create an API key in the organization settings, or add https://mcp.fullreach.ai as a connector in your assistant and approve it.
2. Ask the three questions above. Compare each figure with the dashboard page the tool reads.
3. For a client report, copy the Looker Studio report and give each data source your project. Our guide on reporting to clients gives the five numbers and their order.