Quick start¶

Once mr-review is running (see installation), open the app in your browser:

• All-in-one: http://localhost:17240
• Standard deployment: http://localhost:17242

1. Add an AI provider¶

Go to Settings → AI providers → Add provider.

Fill in:

• Name — a label for this provider (e.g. "Claude Sonnet")
• Type — Claude, OpenAI or OpenAI-compat
• API Key — your provider's API key
• Base URL — leave blank for the backend's own default; for Claude, set it only to go through a gateway such as LiteLLM
• Models — Fetch models from API lists what the endpoint offers with the key you just typed; add the ones you want. The first is the default — Make default changes it.

Examples:

See AI providers for the full field list and what each model accepts in a dispatch.

2. Add a VCS host¶

Go to Settings → Git hosts → Add host.

Fill in:

• Name — a label for this host (e.g. "My GitLab")
• Type — GitLab, GitHub, Gitea, Forgejo or Bitbucket
• Base URL — the instance's web URL (e.g. https://gitlab.com). Bitbucket ignores it
• Access Token — a personal access token; for Bitbucket, username:app_password

Token scopes per host type are on the VCS hosts page.

3. Pick an MR¶

Browse the left panel: select a host, then a project, then click an open MR to load its diff.

4. Run the review pipeline¶

Work through the pipeline stages at the top of the page:

1. BRIEF — choose a review preset, add any custom instructions for the AI
2. DISPATCH — start the AI review; comments stream in as they are generated
3. POLISH — edit, keep, or dismiss individual comments
4. POST — send approved comments back to the merge request on its host

Comments will appear directly on the MR for your team to see.