How Meilleurtaux became the first French broker to launch a mortgage app on ChatGPT
Meilleurtaux is one of the largest credit and mortgage brokers in France. The company helps millions of households compare and secure financing for their home, car, and life projects.
On 29 May 2026, Meilleurtaux became the first broker in France to launch a dedicated mortgage lending app inside ChatGPT. Customers can now run a borrowing-capacity simulation in plain French, using Meilleurtaux's real-time rates from its 125 partner banks, and then choose to be contacted by a Meilleurtaux mortgage expert or to submit their application on the website. Read the full announcement on AFP.
The app is currently available to ChatGPT users in France: open Meilleurtaux on ChatGPT
The integration is part of Meilleurtaux's "AI First" strategy, which sits inside a broader plan to double the size of the group within three years. Internally, the company is already running AI in back-office automation and customer engagement via Salesforce Agentforce at Meilleurtaux Placement. The ChatGPT app is the next step, this time directly facing customers.
This post explains how Meilleurtaux's engineering team and Manufact got from a working prototype to a production-ready ChatGPT app in a matter of weeks, end-to-end in 6 weeks.
The starting point
When we first met Meilleurtaux's team (Damien on product and tech, Emeline on PMO, Dimitri as tech lead), the brief was already sharper than most teams at this stage:
- A crisp use case. The single question every mortgage journey starts with, "how much can I borrow?", turned into a guided conversational flow inside ChatGPT.
- Figma prototypes for the full experience. The conversation, the simulation cards across 15, 20, and 25-year terms, and the lead-capture handoff into the existing Meilleurtaux advisor workflow.
- A clear quality bar. The result had to feel like a Meilleurtaux product.
The collaboration
Meilleurtaux built their MCP App using mcp-use SDK, the fullstack MCP framework to develop MCP Apps for ChatGPT / Claude.
It gave the team:
- A typed server with clean tool definitions.
- Sensible defaults for tool naming, input validation, and authentication. These are the patterns most likely to pass a ChatGPT app review.
- Drop-in support for MCP UI views, which the simulation cards use.
As part of the enterpise plan, Meilleurtaux had access to Manufact's white-glove support. This included hand-on help from our side to build the first version of the MCP App, weekly sync-ups, and live reviews on the parts of the protocol that are easy to get wrong. For example:
- Tool annotations so ChatGPT's permission popup does not break the conversational flow.
- Widget structure so the simulation cards render the same way across ChatGPT clients.
End-to-end, the development and deployment of the MCP App went live in less than 6 weeks.
One-click deploy
Once the MCP App was ready, the team connected the GitHub repository to Manufact Cloud. After that, deployment collapsed to one action: push to main, get a production MCP endpoint.
Hosting
Connect GitHub, get branch previews and a production MCP endpoint with managed TLS, env vars, and deploy history. Promote, roll back, and wire CI without operating any infrastructure.
Every pull request got its own preview environment with a unique MCP URL. The team used those previews the same way they would use Vercel previews for a web app Open a PR, click the URL, test the change end-to-end, then merge. No staging server to maintain. No environment drift.
QA across clients
The biggest blocker the team flagged on call one was "how do our non-developers test this?". Their QA team and product managers cannot spin up a local MCP server every time they need to validate a flow.
Manufact's hosts an MCP Inspector connected to your MCP App for this purpose:
- Inspect every tool call and response in real time.
- Re-run flows with different inputs in the platform.
For Meilleurtaux's QA team, the workflow shifted from "ask a developer to try it on ChatGPT" to "open the Inspector link in the PR and try it."
Cross-client testing
Run the same eval suite across ChatGPT, Claude, and other MCP clients. Catch client-specific regressions before users do.
Publishing checks
Submitting an app to the ChatGPT App Store is opaque for most teams. The list of approval criteria is long: protocol conformance, tool naming, security headers, allowed domains, asset requirements, prompt-injection hygiene. A single missing piece can mean a multi-week rejection cycle.
Before submitting, Meilleurtaux ran Manufact's publishing checks against their production endpoint:
Publishing checks
Audit your MCP app against ChatGPT App Store and Claude Cloud Connectors requirements: protocol, tools, security, domains, assets, and more, directly on Manufact Cloud. Learn more →
The checks surfaced everything that would have triggered a manual review fail. A few tool annotations that needed tightening. Asset dimensions for the store listing. A domain reachability fix. The team resolved them in an afternoon. The app went into the store on the first try.
Live in production, with analytics from day one
The app went live, and the team had visibility from day one through Manufact Analytics:
- Which tools get called most.
- Error rates broken down by failure mode.
- Cross-client traffic split, so the team can see how many users come in via ChatGPT versus other clients like Claude.
The result
| Before Manufact | With Manufact and mcp-use | |
|---|---|---|
| Time to first MCP app live | Open-ended | Under two months |
| QA loop | Developer-blocked | Self-serve via hosted Inspector |
| Submission risk | High, opaque criteria | Pre-flighted via publishing checks |
| Post-launch operations | DIY DevOps | Built-in hosting, analytics, alerting |
| Engineering focus | DevOps and protocol plumbing | The product experience |
| Strategic outcome | n/a | First French broker on ChatGPT |
Meilleurtaux's team has full ownership of the product and the code.
Manufact handled hosting, testing, submission, and monitoring, so the team could spend its time on the product itself.
What's next
The ChatGPT app is one step in a broader plan. Meilleurtaux is now exploring how to bring the same conversational flow into its own properties and how to extend the MCP surface beyond borrowing simulation into the full mortgage journey, and eventually into the rest of the group's business lines.
If you are a financial services team, or any consumer brand, thinking about how to show up inside ChatGPT or Claude, the same approach works:
- Build with mcp-use, the open-source MCP SDK.
- Deploy with Manufact Hosting. Connect your GitHub repo and ship in one click.
- QA across clients with the hosted Inspector.
- De-risk submission with publishing checks.
- Operate with analytics.
Meilleurtaux was founded in 1999. It is one of the largest omnichannel financial services brokerage platforms in France, covering credit, insurance, and investments. Learn more at meilleurtaux.com.
