Guidal
Sign in

MCP setup

Connect Claude Code, Cursor or another MCP-capable editor to Guidal.

The Model Context Protocol (MCP) lets an AI assistant in your editor operate Guidal directly — creating projects, deploying services, reading logs — without you translating intent into commands.

Prerequisites

The MCP server reuses your CLI session, so install and sign in first:

bash
pip install guidal
guidal login

Confirm you are authenticated:

bash
guidal whoami

Register with your editor

bash
guidal mcp setup

This writes the MCP server configuration your editor expects. Restart the editor afterwards so it picks up the new server.

Check the connection:

bash
guidal mcp status

Verify it works

In your editor's assistant, ask something read-only first:

List my Guidal projects.

If the assistant returns your projects, the connection is live. If it says it has no such capability, the editor has not loaded the server — restart it, and confirm guidal mcp status reports the server as configured.

Run the server directly

Editors normally start the server themselves. To run it manually — for debugging, or for an editor you are configuring by hand:

bash
guidal mcp serve

Remove it

bash
guidal mcp remove

How authentication works

The MCP server calls the Guidal API using the credentials stored by guidal login. There is no separate token to manage and nothing to paste into your editor's settings.

That also means the agent can do what you can do, in the organizations and projects you have access to. It does not get broader permissions than your account.

For unattended agents that cannot complete a browser login, export an API key as GUIDAL_ACCESS_TOKEN before starting the server.

Sessions expire

If the assistant starts reporting authentication errors after working fine, your session has expired. Run guidal login again — the MCP server picks up the refreshed credentials without reconfiguration.

What the agent can and cannot do

Delete operations are deliberately not exposed to agents. Removing a service, database or bucket is done by a human through the CLI or dashboard. See MCP tools and workflows for the full boundary and for how the agent discovers what is available.