Skip to content

Letting an AI agent operate your projects, with a preview first

The built-in MCP server lets agents list, deploy and inspect projects. Anything that changes state is previewed and needs a confirmation code.

4 min readREM CloudHub team

MCPAgentsSecurity

AI agents are good at the repetitive parts of operations: reading logs, checking which deployment is live, redeploying after a fix. They are also very confident. REM CloudHub's MCP server is built so an agent can do the work without being able to surprise you.

Same permissions as a person

An agent connects with a token that belongs to a person. It sees the projects that person may see and can do what that person may do, nothing more. If the person loses access, so does the agent.

Read freely, change carefully

Tools that only read (list projects, read logs, inspect a deployment) answer straight away. Tools that change something work in two steps:

  1. The agent calls the tool. RCH answers with a preview of what would happen and a confirmation code.
  2. The agent, or the person watching it, calls again with the code. The code is valid for two minutes and works once.

The preview is the moment to catch a wrong project or a wrong branch, before anything moves.

Every call is in Activity

Each tool call is recorded with the person behind the token, the tool, the arguments and the result. Policy decisions (allowed, refused, and why) are listed too, so you can see what an agent tried to do as well as what it did.

Getting started

Create a personal access token in the console, add the MCP server address to your agent's configuration and ask it to list your projects. The docs have the address and a configuration example for common MCP clients.

Was this useful?

Subscribe to the newsletter

New posts and product news from REM CloudHub, a few times a month. Nothing else.