Log inStart free
← Blog
aimcp

MCP: let your AI agent deploy (safely)

How Aloof's remote MCP server works, what an agent can and can't do, and how to keep a human in the loop for destructive actions.

Aloof ships a Model Context Protocol server at https://mcp.aloof.si/mcp. Any MCP client that supports remote servers with OAuth (Claude, ChatGPT connectors, Cursor and others) can connect. You can also run it locally over stdio.

Authentication

The remote server implements OAuth 2.1 with PKCE and dynamic client registration. The client discovers the authorization server, you log in to Aloof in the browser, and the client receives a scoped token you can revoke from the dashboard. Headless setups can send an API key as a Bearer token instead.

What the agent can do

34 tools cover the platform: deploy_app, deploy_with_config (an aloof.toml as JSON), get_status, app_status, get_logs, scale_app, set_env, set_autoscale, domains, volumes, managed Postgres, releases and rollback, snapshots and off-site restores, private registry and git credentials, deploy tokens and audited exec.

# you: "Deploy github.com/acme/api on Aloof with 512 MB, scale to zero, and tell me the URL."
# the agent: list_plans → deploy_app(name="api", git_url=..., plan="shared-1x-512mb")
           → set_autoscale(min=0) → app_status → https://api.aloof.si/

Keeping it safe

For agents that prefer REST

/llms.txt is a compact, accurate description of the API, and the OpenAPI spec is at /api/docs. Agents with shell access can also just use the single-file CLI.

Try it

Free tier, hard-capped at $1 of usage. No card needed.

More posts