Skip to main content
The Alumio MCP server lets an AI assistant work with your Alumio environment from a normal chat conversation. You can ask Claude or ChatGPT which tasks failed last night, why a run of an incoming configuration stopped, or what a route does, and the assistant reads the answer straight from your environment. With the right permissions, it can also retry failed tasks or stop a stuck job for you. MCP (Model Context Protocol) is an open standard that AI apps use to connect to other systems. Alumio offers it as a remote server, so you only need the address of your environment and your normal Alumio login.
The MCP server is in beta. The available tools and their behaviour may still change.

Server address

Every environment has its own MCP server. Take the address of your environment and add /api/v1/mcp:
The assistant works in one Space at a time. The address above uses the Default Space. To connect to another Space, add its identifier:
Add one connector per Space if you want an assistant to work in more than one Space.

Sign-in and permissions

You connect by signing in with your own Alumio account. When you add the connector, the AI app opens an Alumio page where you check the account, the environment and what the assistant may do, then choose Allow access or Deny. You never have to share a password or API key with the AI app. The assistant acts as you. It can only reach what your role allows, and the access you grant can lower that further: To make changes through the MCP server, your role must be Editor or higher. A Viewer can only read, whatever access is granted. See Users and roles.
Changes made by an assistant affect live integrations. On a production environment, the sign-in page warns you about this. Only grant change access when you need it.
You stay signed in while you use the connector. If you don’t use it for 30 days, the AI app asks you to sign in again. To stop an assistant from using your environment, disconnect or remove the connector in the AI app.

What the MCP server has access to

The MCP server gives the assistant a fixed set of tools. The assistant chooses which tool to use based on your question.

Reading data

Available to every role.

Making changes

Requires the Editor role or higher, and change access on the sign-in page. Both tools work on all tasks or jobs that match a filter, so one request can change many items. They first count how many items would change. Assistants are instructed to show you that number and to ask before they make the change. Your AI app may also ask you to approve each change.

What stays protected

  • Secrets in configurations, such as passwords and authentication headers, are shown as <REDACTED>. Placeholders such as ${PASSWORD} are shown as they are.
  • The values of encrypted environment variables are never shown.
  • The assistant only sees the Space in the server address.
  • The MCP server cannot create, edit or delete configurations, routes or environment variables.

Next steps