API

Connect to an AI assistant

Add GeoReputation to Claude, ChatGPT or Microsoft 365 Copilot so you can ask about your brand's AI visibility in the tool you already use.

GeoReputation runs an MCP server, so an AI assistant can read your visibility data and answer questions about it directly. You ask "why did our score drop last month?" in Claude, and it reads your actual runs to answer.

This needs no code. You paste one URL and sign in.

Connecting requires the API Access add-on on your account. If you do not have it, the connector will add successfully and then every question will come back saying the add-on is missing. An account owner can enable it in Settings.

What you connect to

Server URLhttps://mcp.georeputation.com/mcp
OAuth client IDB5qQAoj0iDdteUVhi6GKnX0imb847Ze6n6iOK7JK
Accessread-only

The client ID is meant to be public and is the same for everyone. It is not a secret and it grants nothing on its own: it identifies Claude to our sign-in, and the data you see still comes from your own account after you sign in. There is no client secret to enter.

The connection is read-only by construction, not by policy. Every request it makes carries your own sign-in and goes through the same public API your own script would use, and that API refuses writes from any token. An assistant connected this way cannot start a scan, change a brand or spend your credits. It also only ever sees the brands your account can see, because the token is pinned to your organisation before any data is read.

Claude

You can do this yourself on Free, Pro, Max and Ultra.

  1. Open Customize > Connectors
  2. Add, then Add custom connector
  3. Enter a name and the server URL above, then Continue
  4. Under Authentication choose Sign in now. If Claude marks the settings as Detected, keep what it found and move on, because our server hands it everything it needs. Only if it asks for an OAuth client do you choose Use your own OAuth client, paste the client ID above, and leave the secret blank.
  5. Add, then Connect, then sign in to GeoReputation

Step 4 asks for less than it used to. Our server now registers a client for your tool automatically, two ways: it publishes an RFC 7591 registration endpoint, and it accepts a client whose identity is its own published metadata document. Either way the tool asks for a client of its own and gets one, so the client ID in the table above is a fallback for a tool that supports neither rather than the normal path. It still works, and it is still public.

On Team and Enterprise an Owner adds it once under Organization settings > Connectors, and each member then clicks Connect and signs in with their own account, so everyone sees only what their own account can see.

Claude cannot change a connector's authentication after it is added. If you pick the wrong option you have to remove the connector and add it again, so it is worth following step 4 exactly.

Free plans are limited to one custom connector.

ChatGPT

Gated by OpenAI rather than by us, and the work belongs to a workspace administrator. Custom MCP servers are a Business and Enterprise or Edu feature on the web, OpenAI's FAQ separately says Pro accounts can connect a custom MCP server with read and fetch permissions, and Plus is named nowhere. We also have to issue an OAuth client for your workspace first, because the client ID above is registered against Claude's callback and will not work anywhere else.

Full setup for a ChatGPT administrator, including the plan table, the click path and a block you can paste into a ticket.

Microsoft 365 Copilot

A route exists and it belongs to your tenant administrator: a Global Administrator or AI Administrator creates a custom federated connector, and every person who uses it needs a Microsoft 365 Copilot licence. That route asks for a client secret registered in your own Teams Developer Portal, so contact us and we will issue credentials for your tenant rather than publish a secret here.

The thing to know before anyone starts: we have not yet watched a tool call arrive from Copilot Chat. Several public reports describe connectors that connect and list their tools correctly while Copilot does not reliably call them, because invocation is decided by Copilot's own orchestration. We would rather say that now than have you find out after an admin has spent an afternoon.

Full setup for a tenant administrator, including the roles, the OAuth fields, and the six conditions that all have to be true before a user sees it.

What you can ask

The server exposes eleven tools, and they are shaped around questions rather than around our database. Ask in your own words; you do not need the tool names.

AskWhat it reads
"What brands can you see?"your brand list with each one's current score
"How visible is Acme in AI answers, and is that up or down?"the score, the change over the window, and the per-assistant breakdown
"Which questions do we win and which do we lose?"your prompt categories, and the prompts inside one
"Why did the score move last month?"the categories that moved most, largest move first
"What should we do about it?"your recommendations, each with why it was raised and the action suggested
"Send me the latest in-depth report"your reports, newest first, with a link to each
"When did you last check us?"your monitoring runs, with what each one scored and whether it finished
"What did each assistant say in that run?"every answer in one run: the model, whether you were mentioned, the sentiment and the rank
"Show me what ChatGPT actually said"one answer in full, word for word
"What changed for that one question?"mentions starting or stopping, rank moves, competitors appearing, sentiment shifts
"Which client accounts can you see?"the organizations this account reaches, for an agency that manages clients

Windows are 7d, 30d or 90d.

If you manage client accounts, cross-account reading is built but not switched on for any connector yet, so today every connection sees one organization: its own. "Which client accounts can you see?" will answer with that single account.

When it is enabled it will work one account at a time, reading your own brands by default and a client's only when you name that client, so a question about one client cannot quietly pull in another's data. Ask us if you need it.

Most accounts have exactly one organization and can ignore this entirely.

Scores run 0 to 1000, not 0 to 100, and they are not percentages. A move is best described in points. The bands the product itself uses, so a number means the same thing here as on your dashboard:

ScoreBand
0Not Visible
1–199Low
200–399Limited
400–599Moderate
600–799Good
800–1000Strong

Two things worth knowing, because they change how you read an answer:

  • A report comes back as a link, not as a file. The assistant will hand you a URL to open. If a report is still being generated it has no link yet, and the assistant will say so rather than tell you it is missing.
  • "Nothing changed" and "nothing to compare against" are different answers. A question we have only measured once has no previous run to compare to. The assistant distinguishes the two, so take the wording literally.

If something does not work

  • Every question says the add-on is missing. The connector is fine and the account needs API Access. An owner can enable it in Settings.
  • You are asked to sign in again repeatedly. The connector was added with the wrong authentication option. Remove it and add it again, following step 4.
  • Answers stop partway through a long question. Each connection is limited to 60 requests a minute. Ask for one brand or one window at a time.
  • Copilot sees the connector but never uses it. Phrase the request closer to the tool descriptions above, naming the brand. If it still does not call it, tell us, because that is the behaviour we are tracking. There is more on this in the Copilot setup page.