The MCP server connects brandflare to an AI assistant, so you can ask about your brands in plain English where you already work: Claude, ChatGPT, Cursor, or any assistant that accepts one. MCP, the Model Context Protocol, is the standard an assistant uses to read from another product.
It's part of the Pro and Agency plans. The assistant reads what the API returns and changes nothing.
What you can ask
Once it's connected, ask the way you'd ask a colleague:
- "How has Northwind Coffee's Flare score moved over the last month?"
- "Which questions are we missing from on ChatGPT?"
- "What is AI getting wrong about us, and which of it matters most?"
- "What should we fix first, and what are the steps?"
- "Which websites does AI cite in our category that never mention us?"
- "Write Monday's update for each of my brands, with what changed."
The assistant fetches the answer from brandflare and replies with it. It can chart the figures, compare brands, or fold them into a report or an email it's drafting for you.
Connect it in two minutes
- Copy the connector address. It's also on Settings › API & MCP:
https://brandflare.io/api/mcp
- Add it to your assistant, as below.
- Sign in to brandflare when the assistant asks, and allow it to connect.
- Ask it "What brands do I have in brandflare?" to see it working.
Claude
Open Settings › Connectors, choose Add custom connector, name it brandflare and paste the address. Select Connect and sign in. It works in Claude on the web, the desktop app and the phone app.
ChatGPT
Custom connectors are added in Settings, under Apps & Connectors, with developer mode switched on. Paste the address, leave authentication on OAuth, and sign in when asked.
Claude Code
Run this once, in a terminal, then use /mcp to sign in:
claude mcp add --transport http brandflare https://brandflare.io/api/mcp
Cursor and other tools
Anything that takes a remote MCP server works with the same address. In Cursor, add this to mcp.json and sign in when it asks:
{
"mcpServers": {
"brandflare": { "url": "https://brandflare.io/api/mcp" }
}
}
Assistants move their settings from time to time. If a menu isn't where this page says, look for "custom connector" or "remote MCP server".
Signing in
Signing in is the same sign-in as brandflare itself, with two-factor if you use it. The assistant never sees your password: it is given a pass that says who you are, and brandflare checks your place on the workspace every time it asks for something.
- Everyone on the team can connect. Each person connects their own assistant with their own sign-in. Every role can, because every role can read the workspace's brands.
- It reads every brand on the workspace. To keep a connection to one brand, use a key instead, as below.
- It ends when their place does. Remove someone from the team and their assistant is refused from its next question. The same goes if the plan ends or the account is closed.
- Two-factor is honoured. If the workspace requires two-factor and someone hasn't set it up in time, their assistant is refused until they do.
What the assistant reads becomes part of your conversation with it, and is handled under that assistant's own terms.
With a key instead
A connection can carry one of your API keys in place of a sign-in. Use a key when:
- It should read one brand only. A key for one brand keeps the assistant to that brand, which is what you want for an assistant a client will use.
- Nobody is there to sign in. A script or a scheduled job has no one to press Allow.
Where a tool lets you set a header, send the key that way:
claude mcp add --transport http brandflare https://brandflare.io/api/mcp --header "Authorization: Bearer YOUR_KEY"
Where only an address can be given, put the key at the end of it:
https://brandflare.io/api/mcp?key=YOUR_KEY
An address with a key in it works for anyone who has it, so share it as carefully as the key itself. Keys are made, named and deleted on Settings › API & MCP, and deleting one stops its connection at once.
What the assistant can read
The server gives the assistant nine tools. It chooses between them by itself, according to what you ask.
| Tool | Returns |
|---|---|
list_brands | Your brands, each with its current Flare score |
get_brand | One brand: score, five factors, last and next check |
get_score_history | The score and factors day by day |
get_engines | How each AI engine treats the brand |
get_competitors | Share of voice against competitors |
get_questions | The brand's questions, and whether each engine's latest answer named it |
get_claims | What AI has said that the fact sheet contradicts, why it's wrong and what's true |
get_recommendations | What to do next, most important first, with the steps |
get_sources | The websites AI cites, the brand's own pages among them, and who to approach |
Each returns what the matching address of the API returns, and that guide explains every field. A brand checked in more than one market can be asked about one market at a time or all together.
The assistant can't read the full text of an answer, and it can't change anything: it can't settle a claim, mark a recommendation done, edit a question or start a check.
How fresh it is
The assistant is told what the checks have measured. A brand's visibility, average position, Flare score and share of voice against competitors are measured every day. Everything else comes from the brand's weekly check and changes once a week. Asking twice in a day returns the same thing.
Limits
A person's assistant can make 60 requests a minute, and so can each key. One question to an assistant usually takes two or three.
When something goes wrong
The assistant is told what is wrong in a sentence, and passes it on.
| What it says | What to do |
|---|---|
| This sign-in isn't on a brandflare workspace | Open brandflare once to finish setting up, or ask your workspace's owner to invite the address you signed in with |
| It is part of the Pro and Agency plans | The workspace isn't on one of them, or its plan has ended. It works again when the plan resumes |
| This account is not active | The account has been closed or suspended. Raise a support request |
| This workspace requires two-factor sign-in | Set up two-factor in brandflare, then ask again |
| This connection's brandflare key isn't recognised | Check the key was copied in full. If it was deleted, make another, or remove it and sign in instead |
| No brand with that ID is available | The connection's key reads one brand, and you asked about another |
| Too many requests | Wait a minute and ask again |
If the assistant asks you to sign in again, the sign-in has run out: sign in and carry on. If it says brandflare couldn't check the sign-in, wait a moment and ask again. If it can't connect at all, check the address ends in /api/mcp. For anything else, raise a support request.
Using it well
- Name the brand. "How is Northwind Coffee doing on ChatGPT?" gets a straighter answer than "how are we doing?" on a workspace with several brands.
- Ask for the change, not only the number. The score a fortnight ago and the history come with every brand, so "what moved, and where?" is a question it can answer.
- Put two things together. "Which of our weakest questions are about pricing, and is anything AI says about our pricing wrong?" draws on the questions and the claims at once, which no single page does.
- Start from the worst. Claims come most serious first and recommendations most important first, so "the top five" is a sound place to begin.
- Treat an early score as provisional. While a brand's score is marked "Early score, still settling", the assistant is told so. Ask it to say so in anything it writes for a client.
- Use it beside Export Report. The assistant is for questions between reports; the report is the document a client keeps.