Deploy From Claude, Cursor, or Codex
Your MapleDeploy server has an MCP endpoint built in. Turn it on, and your AI coding agent can set up a project, add your app and its database, fill in the environment variables, and deploy it, all on your own server in Toronto.
The Model Context Protocol (MCP) is how AI agents like Claude, Cursor, and Codex connect to outside tools. Your server can serve it at /mcp as soon as you turn it on, with no scripts to write and nothing extra to host.
From repo to running app, in one conversation
- 1
Ask in plain words
"Set up a staging project for this repo with a Postgres database, add the env vars from .env.example, and deploy it."
- 2
It creates the project
The client makes a project and a staging environment on your server, the same ones you would click together in Coolify.
- 3
It adds your app and database
It creates the app from your Git repository and a Postgres database beside it. MySQL, Redis, and Coolify's one-click services work the same way.
- 4
It sets your environment variables
It reads what your app expects from the repo and adds each variable to the app, so you are not pasting values into a form.
- 5
It deploys and checks the build
It starts the deploy and reads the build output. If the build fails, it can tell you why, fix the cause, and deploy again.
What your AI agent can do once it's on:
Set up and configure
Create projects, apps, databases, and one-click services, manage environment variables and backup schedules, and clone or move resources between environments.
Deploy and roll back
Deploy after a change, force a clean rebuild, cancel a stuck deploy, roll back to an earlier image, or restart anything.
Debug what broke
Pull logs or a failed deploy's build output into the conversation, and find everything that looks down.
Limited by its token
The API token you give the client sets its limits. A read token can look around and a deploy token can ship. A read:sensitive token can read logs and build output, and a write token can change your setup.
How it works
- 1
Turn it on
In the MapleDeploy dashboard, open your server's Tools tab and turn on the MCP server. You see what it exposes before you confirm.
- 2
Create an API token
In Coolify, go to Keys & Tokens, then API Tokens, and create a token with only the abilities your client needs.
- 3
Connect your client
Add your server's /mcp endpoint and your token to Claude, Cursor, or Codex. Then ask it to set up your app.
What it exposes
Once it is on, any Coolify API token on your instance can reach the endpoint, and the allowed IPs you set in Coolify apply to it too. Give each client the narrowest token that does the job.
The realistic risk is prompt injection: text your client reads that tries to steer what it does next, which matters most for a write token. Logs can contain secrets, a read:sensitive token can read environment variable values, and anything the client reads becomes part of its conversation with its AI provider. You can turn the endpoint off at any time, and the setup guide covers token abilities in detail.
On Canadian infrastructure
Your server, its Coolify instance, and the MCP endpoint all run on a dedicated VM in Toronto. Deploys run there, and your application data lives there, governed by Canadian privacy law rather than the US CLOUD Act.
The endpoint uses your server's existing HTTPS address, so there is no new port to open, and it stays off until you turn it on.
Built on open source
The MCP server is not a MapleDeploy add-on. It ships with Coolify, the open-source platform your server runs, so you can read exactly how the endpoint and each of its tools behave in the source.
If an AI agent helped you build the project, this closes the loop: the same agent that wrote your code can now deploy it, on infrastructure you control.
Ready to deploy with your AI agent?
30 days free on Starter and Pro. Start a server, turn on the MCP server, and deploy a real project from Claude, Cursor, or Codex. The setup guide walks you through connecting your client.