Maple

What Maple MCP server does

Maple MCP server carries visual review feedback from a running frontend preview into an agent workflow. A reviewer can point at an element, drag over an area, or select text, then describe the problem. Maple records what was selected, where it appeared, and relevant reviewer information so the comment can be acted on after a redeploy.

The project is designed around a review loop: mount Maple in a preview, collect comments, let an agent handle them through MCP, and use a CI check to hold the merge while issues remain open. The optional gate can also wait for reviewer approval. The repository is currently in the 0.x series, so public interfaces may change between releases.

How it works

The frontend includes a Maple route and a script in the preview build. A build-time tagger can attach source details to JSX elements, allowing a review comment to identify the file, line, and column associated with an element.

On the agent side, Maple MCP server exposes the review workflow. The agent waits for new comments, reads each comment with its captured context, implements the requested change, and resolves the comment against the resulting commit. A CI check named maple/visual-review fails while comments are unresolved and can optionally require reviewer approval before passing.

Maple does not require a hosted service. For local work, its proxy stores comments as files under .maple/. A preview can also be reviewed through a loopback bridge, although a guest using that mode cannot gate a merge. Team reviews use GitHub Device Flow so comments retain their authorship without placing a secret in the preview.

Setup and configuration

The smallest local entry point is maple review, which proxies a running application and keeps review data in .maple/. To review an existing preview alone, use maple solo <preview-url>; comments are delivered to the local machine through the loopback bridge.

For a team setup, the documented command is maple setup app --owner=acme. The application must include one Maple route and one script in its preview build. The SDK has no default store; without a configured store, its comment endpoints return 404.

Tools and capabilities

Maple MCP server supports these workflow capabilities:

• Receive and monitor new visual review comments.
• Read comment context, including the selected page target and available source location.
• Support comments anchored to elements, screen areas, or text passages.
• Apply a code change through the agent workflow and resolve the related comment against a commit.
• Hold merges with the maple/visual-review CI check while comments remain open.
• Generate requested frontend states such as empty, failing, or large-data responses through the mock package.

The repository also includes a classifier package that can score comments against specificity, actionability, concision, standalone clarity, and placement when a supplied classifier is configured. That scoring is optional rather than a requirement for the MCP workflow.

Limitations and notes

The SDK route does not provide a default data store. Local and solo modes have different collaboration behavior: solo review uses a loopback bridge, and a guest cannot gate a merge. The project is Apache-2.0 licensed, but its 0.x version makes no compatibility promise; breaking changes may be introduced when the maintainers consider them appropriate.

A merge gate only addresses the checks and approval conditions configured for it. Maple is intended for frontend review on deployed previews, not as a general issue tracker or a replacement for unrelated code-review systems.