AI without an MCP server
How many quotes sent this month are still unanswered?
I don't have access to your mailbox. I can explain how to search your email, but I can't read or count your messages.
MCP Server · Development & Hosting
Custom MCP servers: development and hosting
- Kimoun integrates, builds and hosts MCP servers. Based in Guadeloupe (France), the team works remotely with clients across France.
- Free assessment: which of your tools can be connected?
- Our own MCP servers run in production — we use what we sell
- Precise quote within 48 business hours, no commitment
Your business tools, connected to Claude, ChatGPT and Mistral. Kimoun builds your MCP server, hosts it and keeps it maintained. Nothing to install on your side.
MCP Server
Built · hosted
· maintained
Understanding MCP
What is an MCP server?
MCP (Model Context Protocol) is the open standard that lets AI assistants access your tools and your data. An MCP server is the connector: it exposes what you decide — your CMS, your accounting, your documents — to Claude, ChatGPT or Mistral, securely.
Without an MCP server, the AI only knows the public web. With one, it works with your data.
Try the difference
The same question, with and without an MCP server
Pick a scenario: on the left, an AI on its own; on the right, the same AI connected to an MCP server.
Our most popular MCP servers
What can be connected
Four MCP server references — use cases running in production today at Kimoun and for our clients. Not a promise: living proof.
Use case · CMS
AI-driven SPIP CMS
Writing, translation, publishing and media library driven by AI — an MCP server built for a SPIP website publisher.
Use case · Documentation
Obsidian knowledge base
The AI reads, writes and organises the vault. Company memory becomes usable in plain language.
Use case · SEO
SEO data in plain language
Google Trends, Search Console, autocomplete — queried straight from the AI assistant, without opening a dashboard.
Use case · Mailbox
Connected mailbox
Reading, searching, sorting and sending emails, driven by AI. Your mailbox becomes a tool the AI knows how to use.
The offer
Three tiers, from existing connector to custom build
From ready-made connector to custom build: three entry points, delivered remotely to clients across France.
Integration of existing connectors
Hundreds of open-source MCP servers already exist — WordPress, Notion, n8n, GitLab… Kimoun selects the right one, deploys it, secures it (authentication, filtering of exposed tools), customises it and hosts it. The ideal entry point for an SMB.
Custom development
Audit of your tools and APIs, definition of the tools and resources to expose, development, testing with your team. An MCP server designed for your data and your workflows — not a generic connector. Often the natural next step of an AI and automation project.
Managed hosting, across both
French infrastructure (OVH), Docker containers isolated per client, Git-versioned deployments, S3 backups, monitoring. Maintenance follows the MCP protocol as it evolves — an unmaintained server stops working. From €49/month.
Published pricing
Public prices, transparency by design
Few providers publish their MCP server prices. Here are ours. Every project gets a precise quote, free of charge and with no commitment.
The entry point
Integration of an existing MCP connector
from €490 ex. VAT
An open-source MCP server, deployed, secured and hosted for you
Your tool is widely used (WordPress, Notion, n8n, GitLab…)? An open-source MCP server probably already exists. Kimoun selects it, deploys it, secures it — authentication, filtering of exposed tools — customises it and hosts it. Nothing to install on your side.
€900 – €1,500 ex. VAT with advanced authentication or adaptations
Ce qui est inclus
- Selection of the open-source connector suited to your tool
- Deployment on Kimoun's managed infrastructure (France)
- Security: authentication, filtering of exposed tools
- Customisation to match your workflows
- Works with Claude, ChatGPT and Mistral
Idéal pour : SMBs that want to connect a widely used tool to their AI assistant, without custom development
— or depending on your needs —
Custom development — read-only
€2,500 – €4,500 ex. VAT
The AI reads your business data, safely
Audit of your tools and APIs, definition of the tools and resources to expose read-only, development, testing with your team. Your AI assistant consults your data without ever changing it.
Custom development — auth + write
€5,000 – €9,000 ex. VAT
The AI acts in your tools, under your control
Advanced authentication and write actions: the AI assistant creates, edits and publishes in your tools — within the exact limits you define. Built, tested and validated with your team. If you wish, every AI action waits for human approval before it runs.
Multi-system: quoted from €10,000 ex. VAT
Managed hosting
from €49 ex. VAT / month
Your MCP server runs, Kimoun takes care of it — for the price of two AI subscriptions
French infrastructure (OVH), Docker containers isolated per client, Git-versioned deployments, S3 backups, monitoring. Maintenance included, in step with the MCP protocol as it evolves — an unmaintained server stops working.
€149 ex. VAT / month for the business tier — beyond that, on quote
Why Kimoun
Managed MCP server hosting — instead of self-hosting
Self-hosting an MCP server is entirely possible — and local remains the right place to test. What you buy from Kimoun is not a machine: it is a publisher’s and managed-services craft — the same one behind our managed web hosting. Git-versioned, reproducible deployments, continuous integration, rollback, monitoring, security, tested backups.
A self-hosted server works on the day it is installed. Who keeps it running six months later, once the protocol has changed three times? This is not theoretical: the MCP specification shipped another evolution in July 2026. Maintenance in step with MCP’s evolution is included in managed hosting. And if you prefer your own infrastructure, Kimoun can deliver on your servers, with a managed-services contract.
An MCP server is only worth something if it still runs six months from now. That is exactly Kimoun’s craft: hosting it, monitoring it and maintaining it.
MCP server FAQ
The questions we get about MCP servers
Straight answers to the questions we hear before an MCP server project starts.
An MCP server is the connector that gives an AI assistant — Claude, ChatGPT or Mistral — secure access to your tools and your data. MCP (Model Context Protocol) is the open standard behind that connection: the server exposes only what you decide — your CMS, your accounting, your documents — and nothing else. Without an MCP server, an AI assistant only knows the public web and whatever you paste into it. With one, it works directly with your data, up to date, with no retyping. In French, the same thing is called a « serveur MCP » — but even in France, MCP server is the term most people search for.
An MCP server, an API and a custom GPT solve different problems. An API is a tool’s technical front door: for an AI assistant to use it, you need a specific integration, built assistant by assistant. A custom GPT lives with a single provider: it tailors ChatGPT with your instructions and documents, but none of it is reusable from Claude or Mistral. An MCP server sits in between: it translates your existing APIs into an open standard that every compatible assistant understands — the same connector serves Claude, ChatGPT and Mistral, with no rewriting. In practice, the three combine: the MCP server builds on your APIs and replaces the integrations you would otherwise rebuild for each assistant.
A simple example of an MCP server: the mailbox. The server exposes four functions — read, search, sort, send — and your mailbox becomes a tool the AI knows how to use: “find the quote we sent that client in March”, “draft a reply to this message”. At Kimoun, this server runs in production every day, alongside three others: an Obsidian knowledge base the AI reads, writes and organises, SEO data (Google Trends, Search Console) queried in plain language, and a SPIP CMS whose writing, translation and publishing are driven by AI. A good MCP server almost always starts this way: one tool, a few functions, a clear scope.
There are two ways to build an MCP server. First: your tool is widely used (WordPress, Notion, n8n, GitLab…) — an open-source MCP server probably already exists among the hundreds published. The work is then to select it, deploy it, secure it (authentication, filtering of the exposed functions) and customise it: that is integration, the ideal entry point for an SMB. Second: your tool is specific — you build custom. The Kimoun approach: audit of your tools and their APIs, definition of the tools and resources to expose, development, then testing with your team. In both cases, the server then has to be hosted and maintained: the MCP protocol evolves fast, and an unmaintained server stops working.
Integrating an existing connector (WordPress, Notion, n8n…) starts at €490. A custom read-only server: €2,500 to €4,500. With authentication and write actions: €5,000 to €9,000. Multi-system projects: quoted from €10,000. Managed hosting, protocol maintenance included, starts at €49/month — while MCP server maintenance alone commonly sells for $200 to $800 per month on the English-speaking market. The initial assessment is free. These prices are public and published — rare in this market — and every project gets a precise quote, free of charge and with no commitment, within 48 business hours.
The MCP server is hosted in France, on OVH infrastructure, in a Docker container isolated per client: your data shares nothing with another client’s. Deployments are versioned with Git, backups go to S3 storage, and monitoring watches the service continuously. On the access side, the MCP server principle is simple: it only exposes what you decide to expose — this tool, that data, read-only or with write access — and authentication controls who connects. Your data is not used to train models. If you prefer your own infrastructure, delivery on your servers is possible, with a managed-services contract.
Yes, an MCP server works with Claude, ChatGPT and Mistral. MCP (Model Context Protocol) is an open standard, published by Anthropic in late 2024 and since adopted by OpenAI and Mistral: the same server works with all three assistants, with no rewriting. In practice, you are not locked into one AI provider — if your team switches from ChatGPT to Claude tomorrow, your MCP server follows, as is. It is also insurance for the future: tools that adopt the protocol today remain usable by tomorrow’s assistants, as long as the server is maintained as the standard evolves. At Kimoun, that maintenance is included in managed hosting.
Take the next step
Let’s talk about your tools — the assessment is free
A 20-minute conversation is enough to identify which of your tools can be connected and where to start. Free assessment, precise quote within 48 h, zero commitment. Oliver answers you directly.