Your agent reads. A human approves.
Your agent prepares a costed plan. Someone on your team approves it, and the Fransys engine executes it.
In this example, you are the one approving.
Stand up the preprod
Prepared by Claude Code, through the Fransys MCP
- CreateLoad balancerpreprod.your-app.com+ €20
- CreateFront1 replica · 2 vCPU · 4 GB+ €24
- CreateAPI2 replicas · 4 vCPU · 8 GB+ €96
- CreateWorkers1 replica · 4 vCPU · 8 GB+ €48
- CreatePostgres1 node · 8 vCPU · 32 GB · 200 GB+ €164
- CreateRedis1 node · 1 vCPU · 6 GB+ €24
Estimated by your agent+ €376 a month
After€2,733 of €3,000
Your turn: approve or reject the plan.
Engine log · no AI
- Waiting for an approved plan.
The airlock, in four steps
Nothing goes from the agent to your production without passing through the airlock. Reads go straight through. Every write waits for approval from someone on your team.
- 01
The agent reads
Logs, metrics, history, costs. Read-only, no waiting.
- 02
It prepares a plan
Costed block by block, before touching anything.
- 03
You approve
Someone on your team. Without approval, nothing executes.
- 04
The engine executes
A tested program: the same action always gives the same result.
Your tools, connected through the Fransys MCP
Claude Code, Cursor, Codex or any MCP-compatible agent connects with one command. It then works on what Fransys has been running in production since 2025: networking, backups, scaling and rollback.
- Claude Code
- Cursor
- Codex
- Any MCP-compatible agent
What you can ask it
Why did the API slow down last night?
Read logs, metrics, history
How much is the preprod costing us this month?
Read costs resource by resource
Deploy the billing branch to preprod.
Act a plan, then a human approval
One line to plug it in
$ claude mcp add --transport http fransys https://mcp.fransys.cloud/mcp
✓ fransys connected · read, and act on approval
$ claude "how is prod doing?"
✓ 6 blocks live · €2,357 of €3,000 this month
What the agent is not allowed to do
The agent reads everything it needs to help you. It never writes on its own, and it executes nothing.
-
Read
Allowed
Logs, metrics, deployment history, costs. No waiting.
-
Prepare a plan
Allowed
Costed block by block, before touching anything.
-
Write to your production
Never on its own
The plan waits for approval from someone on your team.
-
Execute
No
The Fransys engine executes, under your ceiling in euros.
It has already happened
In July 2025, Replit's coding agent erased a project's production database, despite the instruction to change nothing without approval. With Fransys, that plan stops at the airlock and waits for a human.
Source: OECD, AI Incidents
Play this plan in the exampleHave your agent check Fransys
Paste this prompt into your agent, at the root of your project. It reads our technical sheet, written for it, and answers you with your numbers.
On a computer, your agent does it on its own: it reads our technical sheet, written for it, and answers you with your numbers. From your phone, the simplest way is to open the console, for free.
Read https://www.fransys.cloud/en/for-an-agent, then look at our repository. Tell me if Fransys can host this project, what would need to change, and what it would cost per month.
Start for freeYour questions
I can do it quickly with an AI. Why Fransys?
An agent is fast, including when it is wrong. It writes a deployment in an hour. Keeping it in production takes backups, scaling, rollback, and an approval before each write, what we call the airlock. Fransys provides all of that, and your agent plugs into it.
What is the agent allowed to do?
It freely reads logs, metrics, history and costs. To act, it prepares a costed plan, which someone on your team approves before the Fransys engine, a program with no AI, executes it.
Do you migrate our current cluster?
No. The platform does not take over a cluster already in place: it deploys from your repository, onto a new cluster. What our DevOps as a service covers.
No credit card, no commitment.