Protocol
- Model Context Protocol
- MCP TypeScript SDK
- JSON Schema
Open your product to AI agents
I build MCP servers that let AI agents such as Claude and ChatGPT use your application, fully remote. Each tool gets scoped permissions, tests and documentation before it goes live.
Jean-Baptiste Thery
Senior full-stack TypeScript & AI freelance developer
Agentic developer, within your rules
Available for fully remote freelance
- Starting point
- Scoping of the tasks agents should perform
- Pricing
- Fixed price, or day rate of €550 excl. VAT (indicative)
- Start date
- To be agreed
001
001·The service
Decide what agents may read, do and ask
AI assistants increasingly act for their users, and the Model Context Protocol (MCP) is the open standard that lets them call your product's functions as tools. Your MCP server decides what an agent may read, what it may change and what it must leave to a person.
I start from the tasks agents should perform, map them to your existing API, then build, test and document each tool.
What the engagement covers
- A map of the tasks agents should perform and the API functions behind them
- An MCP server that exposes those functions as tools, run locally or hosted
- Authentication of each client, with your API keys kept on the server
- Limits on what each tool may change, with confirmation for sensitive actions
- Automated tests of every tool before it goes live
- Documentation and example set-ups for the MCP clients your users rely on
002
002·How it works
From your first message to the handover
The same four steps on every engagement, so you know what happens next and what you receive.
You describe your project
Send a short brief on Malt or LinkedIn: your project, budget and target date. I confirm availability and the next step.
We agree the scope
Scope, schedule, pricing (fixed price or day rate), working hours and how we communicate, agreed in writing before work starts.
You follow the work
Code reviews, demos or written reports keep progress visible. Every change I deliver is reviewed and tested.
Handover and follow-up
Documentation, recommendations and, for development work, code whose rights the contract assigns to you. Maintenance by agreement.
003
003·Projects
MCP and agent tooling I have built
An open-source MCP server and the private gateway behind the product I co-founded.
Open-source MCP server for cited search
Four read-only tools, scoped to one project, give Claude Code and Codex cited passages from local code and documents. One command sets them up.
MCP gateway for a product in production
Coding agents read store, subscription and content data through a hosted gateway. Provider keys never left the server, and every write waited for a preview and my approval.
004
004·Recommendations
What clients and colleagues say
“I strongly recommend working with JB. He was instrumental in getting our frontend off the ground, handling everything with zeal, unwavering dedication, and unparalleled reliability.”
LinkedIn“Je travaille avec Jean-Baptiste depuis quelques semaines et c'est un réel plaisir. Il s'investit fortement, fournit un travail de qualité, et ce, malgré une situation d'urgence sur le projet. Très bonne communication et à l'écoute. Je ne peux que recommander Jean-Baptiste pour vos projets et travaillerai de nouveau avec lui dès que je le pourrai !”
Malt“Excellent niveau, s'acclimate vite avec son nouvel environnement de travail (compréhension de la stack, des outils de communication, de l'industrie), très autonome et même pro-actif sur la prise de décision. Excellente communication du travail en cours et restant. Se rend facilement disponible. Je recommande pour tout dev d'application web complexe et équipe haut niveau.”Malt005
005·Technologies
What an MCP project draws on
006
006·About
APIs first, then MCP servers
An MCP server is an API with a new kind of client. I have built APIs since my first Symfony backends, years before AI agents, and I treat agents like any API client, with the smallest set of permissions their task needs.
Through jCode, I work with public institutions, established companies and startups. My own products, WorkoutGen and Ragmir, taught me what follows a release.
- of web development
- 10 years
- missions completed on Malt (November 2026)
- 80+
- as a freelancer
- 7 years
- from 47 client ratings on Malt (November 2026)
- 5.0/5
Areas of expertise
007
007·Related services
Other ways to work with AI
AI featuresAdd AI, agents or automation to your applicationTechnical consultingGet advice on architecture and AI choicesOpen-source toolExplore Ragmir
How I work with AI on client code is set out on the agentic development page.
008
008·FAQ
MCP servers for your product
What is an MCP server?
A server that exposes some of your product's functions as tools an AI client can call, following the open Model Context Protocol. The agent asks; your server checks who is asking, decides what is allowed and returns the result.
Which AI clients can use it?
Any client that supports MCP, such as Claude and Claude Code, ChatGPT, Cursor or Codex. Support for some protocol features still varies between clients, so we test with the ones your users rely on.
Do we need to rebuild our API?
No. The MCP server sits on top of your existing API and reuses its access rules; scoping shows which functions agents need and which should stay out of their reach.
How do you limit the risks?
Each client authenticates, your API keys stay on the server, inputs are validated and agents get only the tools their task needs. Write actions wait for a preview and an explicit confirmation, as on the gateway I built for my own products.
How much does it cost?
A first set of tools can be quoted at a fixed price after scoping; open-ended work is billed at an indicative €550 excluding VAT per day, as of October 2026.
009
009·Contact
Let's talk about your project.
Tell me what you want to build or improve, your budget and your target date.
- Fully remote
- English or French
- France · UK · US