reviewmy / MCP

The review ends in a diff, not a document.

Findings and client pins become one change brief. Claude Code reads the outstanding requests over MCP, pulls the brief, applies the edits and marks them done, without you copying anything between two windows.

# one token for the account, then name the project per call
claude mcp add reviewmy --token rmy_live_9f2c…

# then, from inside Claude Code
> list the open reviewmy requests
> pull the change brief for #14 and apply it
> mark #14 done

What the agent can do

Five verbs, and nothing destructive.

The server exposes a deliberately small surface. It can read what is outstanding and close the loop; it cannot delete a project, revoke a link, or touch billing.

01

list_requests

Every open finding and client pin on the project, with severity, status, and the file and line where one exists.

02

get_change_brief

The change brief for one request or for the whole queue, what to change, where, and what the reviewer actually said.

03

mark_request_done

Moves a request to resolved and triggers the confirm-the-fix step for whoever raised it. The client still gets the last word.

  • list_projects and get_change_history

    The other two verbs, both read-only. One names the projects on the account, the other shows what has already been changed and when.

  • Scope

    A token belongs to the account, not to one project, and reaches every active project on it. The project argument on each call chooses between them, it does not fence them off. Issue a token per machine you trust, not per client site.

  • Rotation

    Tokens are shown once, hashed at rest, and revocable from the dashboard. Revoking takes effect on the next call, not the next hour.

  • Without MCP

    Every brief has a copy button. The manual path is one paste and is not a lesser version of the feature, it is the same brief.

  • Availability

    Included from Cracked Coder upward. Vibe Coder gets the brief and the copy button, which is the same output by hand.

The loop

Paste a domain. Hand the result to the agent that built it.

Nothing to install on the site under review, and nothing for you to deploy first.

1

Review runs

Seven lenses over the page, plus the code pass if a repository is connected.72/100 · 8 findings

2

Client pins the rest

Their feedback joins the same queue as the AI findings. One list, whatever raised the item.+4 pins

3

Agent closes them

Claude reads the queue, applies the edits, and marks each one done from its side.12 resolved

Stop copying findings between two windows.

One token, one project, and the agent picks up where the review left off.