The CLI and MCP server

Work on your documentation from Claude Code or another agent. Edits arrive in Workover as tracked suggestions you review.

Installing

npm i -g @workover/cli
workover login

login prints a short code, opens your browser, and you approve the connection while signed in to Workover. The token is stored at ~/.workover/config.json, owner-only, and can be revoked any time in Settings → Developer.

Then connect the MCP server to your agent:

claude mcp add workover -- npx -y @workover/mcp

What a connected tool can do

Search and read your documents.
List the projects you can reach.
Create a document, with its body arriving as suggestions.
Propose a revision to an existing document — again as suggestions.
Upload an image into a document, for a screenshot it captured.

An agent proposes; a person accepts. That is the whole model, and it is why connecting one is safe: nothing it does lands in a document without you looking at the diff.

Scopes

A token holds only what you granted when you signed in:

read and propose — always. This is what plain workover login gives you.
publish — only if you asked for it: workover login --publish. It lets the tool publish to your connected WordPress sites.

Publish is opt-in because it is the one thing that changes something outside Workover. Even with it, a token cannot force a push past a “WordPress is newer” conflict — deciding the other side’s change does not matter needs a person who can go and look.

Nothing reaches your account security, your billing or your other tokens, at any scope.

Revoking

workover logout revokes that device’s token. Settings → Developer lists every token with when it was last used; revoke anything you do not recognise. Revocation is immediate.

Where this is useful

Documentation that lives next to code. An agent that has just changed an API can read the affected page, propose the update, and leave it for you — which is a great deal better than the page going stale because updating it was a separate chore.