Guidal
Sign in

Deploy with an AI agent

Let an AI coding assistant build, configure and deploy on Guidal from inside your editor.

Guidal exposes its platform to AI agents through the Model Context Protocol (MCP). Once connected, an assistant in your editor can create projects, deploy services, attach databases and read logs — without you leaving the editor or translating intent into commands yourself.

This is the path Guidal is built for. The dashboard and CLI both assume a human driving; MCP assumes a capable agent that can decide which operations to run.

How it works

Your editor's AI assistant talks to a local MCP server, which calls the Guidal API on your behalf using your existing credentials. The agent discovers what it can do by asking the server, so it does not need the platform memorised.

Set it up

Install the CLI and sign in first — the MCP server reuses that session, so there is no second credential to manage:

bash
pip install guidal
guidal login

Then register the server with your editor:

bash
guidal mcp setup

Confirm it is connected:

bash
guidal mcp status

Editor-specific configuration is in MCP setup.

What to ask for

Once connected, describe the outcome rather than the steps. Useful prompts look like:

  • "Deploy this repository to Guidal as a web service on port 3000, in a new project called checkout."
  • "Create a Postgres database for this project and wire its connection string into the api service."
  • "The api service is returning 500s — read the last 200 log lines and tell me what is failing."
  • "Scale the worker service to 3 replicas and show me memory usage after."

The agent will call the operations it needs and report back. You stay in the loop for anything destructive.

Destructive operations are deliberately restricted

The MCP server does not expose delete operations to agents. Removing a service, database or bucket is done by a human through the CLI or dashboard. This is a deliberate boundary, not a gap — an agent that misreads a request should not be able to drop your database.

Giving the agent context

Every page on this documentation site is available as raw markdown, and the whole corpus can be fetched in a single request. When an agent needs to understand a Guidal concept, point it at:

text
https://guidal.ai/llms.txt

That file indexes every page with a one-line summary and a direct markdown URL, so an agent can fetch only what it needs. For the entire corpus in one request, use https://guidal.ai/llms-full.txt.

Any page on this site can also be copied as clean markdown with the Copy for AI button at the top of the page. See AI access for the full contract.

Working in CI

Agents running unattended in CI should authenticate with an API key rather than an interactive login. Create one with guidal api-keys create, then export it as GUIDAL_ACCESS_TOKEN before invoking the agent. See API keys.