Using MCP to Manage App Reviews From Claude Desktop
2026-07-245 min read
The reviewdesk-mcp package exposes ReviewDesk as a set of tools that an AI host — Claude Desktop, Cursor, Zed — can call directly. Instead of switching to a browser tab to check reviews, you can triage and reply from wherever you're already working.
What's actually exposed
| Tool | What it does |
|---|---|
| list_apps | Lists your connected apps |
| list_reviews / get_review | Lists or fetches reviews, with the same filters as the dashboard |
| draft_from_screenshot | OCR + fuzzy-match a pasted screenshot, then draft a reply |
| draft_from_review | Draft a reply for a review by ID |
| confirm_post | Send, edit, or cancel a pending draft |
| metrics | Pull dashboard-style metrics into your chat |
A typical session
Ask "what's unanswered on the Fintech app this week" and list_reviews returns the filtered set right in your chat. Ask for a draft on a specific one and draft_from_review generates it, sized to the store's character limit automatically. Approve it and confirm_post sends it through the official API — the same status machine (Draft → Posting → Live/Pending/Failed) applies whether you replied from the dashboard, a screenshot, or here.
Nothing about the safety model changes because you're in an editor instead of a browser. Every draft still sits in pending_drafts until you explicitly confirm it.
Why this matters more for agencies
If you're managing reviews for five or fifty client apps, the dashboard is still the right tool for weekly triage. But the MCP server is what makes ad hoc review handling — a client Slacks you about a bad review mid-meeting, you need to check and reply in under a minute — actually fast, because the tool comes to wherever you're already working instead of the other way around.
Chat channels for Discord, Telegram, and WhatsApp ride the same agent service and are specced for a future phase — not available yet, but built on the same foundation.