Skip to content

Automate Prism

You can drive Prism from scripts, CI, AI assistants and other services. This page tells you which one to use, and what none of them can do.

You want to Use Page
Call Prism from any language the HTTP API at /api/v1 with a personal access token API and tokens
Write a Python script or run a terminal command the kicad-prism package and its prism command Python SDK and CLI
Let an AI assistant read designs and leave review comments the MCP server at /mcp AI assistants (MCP)
Let an admin’s assistant look into a slow or stuck server the ops tools and /api/v1/ops Ops diagnostics
Tell another system when something happens in Prism outbound webhooks Webhooks
Link to one exact view of a design view-state URL parameters and window.prism Links to a view
Review every pull or merge request the GitHub Actions and GitLab CI templates Design review in CI

The API, the SDK, the CLI and the MCP server all run through the same API routes as the web app. They apply the same role checks and the same folder visibility. A script or an assistant can never see a project, or make a change, that its owner couldn’t in the browser.

A token or an assistant always acts as a person. It uses that person’s role as it is now, so when the person is demoted or removed, the token loses that access at once.

An AI assistant connected over MCP never signs anything off. Prism refuses these to an MCP client whatever it was granted:

  • approving or publishing a release;
  • catalog QA (accepting or rejecting a component);
  • admin changes;
  • resolving or reopening a comment.

For those steps, the assistant hands you a link and you finish the step in Prism.

No token of any kind resolves or reopens a comment thread. Only a person signed in to Prism can.

A personal access token can carry the release:approve, catalog:qa and admin scopes, so a script can approve or publish a release, perform catalog QA or make admin changes for its owner. Grant those scopes only to a token that needs them. See Scopes.

Prism does not run your scripts for you. There is no scheduler or hosted runner; run scripts from your own machine, CI or server.

The MCP server, ops diagnostics, event retention, token lifetime and webhook targets are set by an administrator. Each setting is described in Configuration.