Managing your product from ChatGPT or Claude: the MCP connector explained
Figue Studio's MCP connector lets your AI assistant (Claude, ChatGPT or a compatible assistant) drop your requests in your own words, read your queue and answer what is waiting on you. For Figue Unlimited clients: what it does, what it does not, and how to connect it.
You are driving, or between two meetings, and an idea comes: “customers should get an email when their order ships”. Usually it ends up in a note, then nowhere. With the connector, you tell your assistant, it turns the idea into a request in your Figue queue, and you get the link. It waits in your queue, in your own words.
MCP, for Model Context Protocol, is an open standard that lets an AI assistant (Claude, ChatGPT or any compatible assistant) use an external tool: read data, create items, ask questions. Figue Studio exposes its functions through this standard. The result: your usual assistant becomes a door to your development team, with no form and no switching tools.
The server exposes the product's functions directly: your assistant's model picks the tool, and Figue Studio runs it. Here is what that allows today.
01
Drop a request
You describe the need, the assistant creates the request in your queue with your words, without rewriting them into jargon, and hands you its link.
02
Know where the queue stands
“What is in progress?”, “Where is the invoice export?”: the assistant reads your board, what is waiting, what is in production, what is delivered and awaiting approval.
03
Answer pending questions
When the team needs a detail, the request moves to “waiting on you”. On every exchange, the assistant flags these and asks you the question; your answer is recorded on the right request, and the queue moves again.
04
Clarify, comment, prioritise
Add a detail, fix a title, tell the team “keep the current logo”, move a request to the top of the queue: all of it happens in the conversation.
05
Ask for an analysis
On a vague request, you can trigger the estimation: an agent reads the need, prepares the scoping and asks its questions. The assistant hands you the result when it is ready. It is an analysis, not a turnaround promise.
03
An example, from voice note to delivery
Here is an example exchange between a client and their assistant over one week. What the client says, and what happens on Figue's side.
One week, four sentences
When
What you say
What happens
Monday
I spend too much time sending invoices.
The assistant asks two questions (which tool, which format), creates the request “Automatic invoice sending”, hands you the link. It enters the queue.
Tuesday
Move invoice export to the top.
The request goes to the top of the queue. The one in progress finishes, this one starts next. Nobody renegotiates.
Thursday
What is blocking the invoice export?
The assistant reads the request: it is “waiting on you”, the team needs access to your invoicing software. You answer in the conversation, the answer is recorded, the queue moves again.
Friday
Tell the team: keep our current logo.
A comment lands on the relevant request. The delivery goes to a preview; you approve from Figue Studio, or ask for an adjustment.
None of these sentences required opening a tool. All of them left a trace in the right place, with a link. That is the difference between a conversation and a queue.
04
What does the assistant not do?
It has to be said clearly, because it is the first question we get: no, your assistant does not build your product, and no, it does not decide anything for you.
It does not code: development is still done by engineers, with Figue's AI to speed things up and a human review on every delivery. It does not promise a turnaround or a size: the server forbids it from announcing a time or an estimate the product has not returned. It does not pick your workspace at random if you have several. It does not ask ten questions at once: one at a time, using the one the team actually asked.
“Never rewrite a request into technical vocabulary before dropping it: carry the person's words. Never promise a turnaround or announce a size the product has not returned.”
05
How do you connect it?
Anyone with a Figue Studio account can do it, without the team's help.
01
Open Figue Studio settings
In your workspace, the MCP section holds the connector address and a step-by-step guide for Claude and ChatGPT.
02
Add the connector in your assistant
In Claude or ChatGPT, add a connector with the address provided. MCP-compatible assistants work the same way.
03
Sign in with your Figue account
A window asks you to authorise access. It is your account, with your permissions; nothing is shared beyond your workspace.
04
Ask a first question
“What is waiting on me?” is a good start. Then drop a request, in your own words, and watch the link arrive.
Because the best request is the one you never write. It comes in the shower, in the car, walking out of a client meeting. By the time you find the tool, it has lost half its context, or vanished. At Figue, the queue is the product: whatever does not enter it does not exist for the team.
We first tried going through an intermediate agent on Figue's side, which redid the tool choice your assistant had just made. Every “list my requests” paid for two reasonings. We removed it: the server exposes its functions directly, your assistant calls them, Figue Studio runs them. What deserves a model, the analysis of a request, stays delegated to a dedicated agent working in its own sandbox.
The connector does not replace Figue Studio, it opens it. Approving a preview, inviting a colleague, booking a call: that stays in the workspace. Dropping, tracking, answering: that can now happen from wherever you already are.
Frequently asked questions
The MCP connector, in questions
What we get asked when connecting it.
Which assistants does it work with?
Claude and ChatGPT, and any assistant that supports MCP connectors. The guide in Figue Studio settings covers the first two; the others follow the same principle: an address, an authorisation with your account.
Will the assistant create requests I did not want?
No. It creates a request when you ask it to, and hands you the link. You can ask it whether a similar topic is already in the queue. You remain the person who decides what enters the queue.
Can I use it by voice?
Yes, if your assistant supports voice: you dictate, it shapes, it drops. It is handy when you travel a lot.
Does my data leave Figue?
The assistant reads and writes in your Figue Studio workspace with your permissions, as you would in the interface. What you then do with it in the conversation depends on the assistant you use and its terms; we add nothing to that.
Do I need to be a subscriber to use it?
The connector is included in Figue Unlimited and available to anyone with a Figue Studio account. It does nothing without a queue behind it: the team is what delivers the requests it drops.
