Capture anything visual
Ask your agent to open a review from screenshots, a page URL (desktop + mobile), PDF slides, or email HTML — each type gets its own checklist and vision lens.
ReviseMy is human-in-the-loop design review for agents. Drop in a screenshot, URL, PDF, or email HTML, mark what matters, and send clear next steps back over MCP — without leaving your AI chat.
01 How it works
From capture to approval, ReviseMy keeps reviews, marks, guest feedback, and lifecycle together — so nothing gets lost between passes.
Ask your agent to open a review from screenshots, a page URL (desktop + mobile), PDF slides, or email HTML — each type gets its own checklist and vision lens.
Optional hints can land first — checklist immediately, optional Claude or OpenAI vision when a key is set. Useful suggestions, never decisions.
Learn moreOpen the review link, point or outline the exact area, set must-fix / nice to have / question / keep. Add suggested copy or answer a question — threads and answers go straight into the agent’s work packet.
Share a private guest link when you want another set of eyes — no accounts. Your marks stay authoritative.
Learn moreTrack marks from open → resolved → verified. When the agent resolves a batch, the review focuses on verify — with before/after evidence and a pass ledger across rounds.
Learn moreApprove or request changes. Structured next_action returns over MCP — and Recent reviews keeps the same token memory humans and agents share.
Humans mark and approve. Agents ship the next pass. Same loop whether you’re reviewing, designing, building, or signing off.
Pick the artifact — each type gets its own checklist and vision lens.
Paste one into your agent — each review takes exactly one source, then the same mark → fix → approve loop.
One MCP endpoint. Paste the config into the host you already use.
Setup, inline review, CI gates, and fair comparisons — when you need the longer path.
02 For agents
create_review accepts images, a capture URL, PDF, or HTML. When you finish,
get_review returns work packets and one clear next_action: wait, apply marks, open another pass, or stop.
In MCP Apps hosts, the review renders inline in chat; CLI hosts still use the review_url link.
second_opinion and guest findings stay suggestions until you accept themresolve_marks can close marks with notes and optional before/after evidencecreate_review + parent_iddesign_checkup_loop can guide the full cycle03 Setup
Choose the app you already use. Try free — no account required — and connect ChatGPT, Claude, Copilot, Cursor, or Grok.
After you have a token, open
Recent reviews
for the same list your agent sees via list_reviews.
Any MCP client can use the same URL and Bearer token; see
Connectors
for the full host list.
04 Credits
Grab a try token — no ReviseMy account. You get 20 credits each month (no rollover), full capture quality. Paid plans are paused for now.
Try
$0
20 credits each month — no rollover.
Reviews stick around 7 days.
05 FAQ
Short answers on hosts, links, marks, passes, billing, and sharing.
You always get a review_url. Open it, mark, approve — works everywhere. On
MCP Apps
hosts (Claude web/Desktop, Copilot, …) the same review can also show up inline in chat. Cursor, Claude Code, and Grok are link-only for now — same loop, just in a tab.
Nope. Grab a try token in Try with your agent — no ReviseMy account. See Credits — 20/mo free while we figure out pricing.
New checkups pause until your monthly pack refills (20 credits, no rollover). Burn is images/PDF = 1, HTML = 3, live URL = 5. Your agent can check the refill date with get_billing. Paid plans are paused while we figure out pricing.
You. M# marks are the brief — that’s what the agent should fix. S# second opinion and G# guest notes stay suggestions until you accept them.
For the free checklist, no — it runs on every upload. For vision hints that mark regions on the capture, add your own ANTHROPIC_API_KEY or OPENAI_API_KEY on the server (your deploy, your usage). Optional: REVISEMY_VISION_PROVIDER to pick Claude vs OpenAI, custom models, or REVISEMY_OPENAI_BASE_URL for Ollama and other OpenAI-compatible endpoints. Then Refresh second opinion on the review. Still suggestions — your marks win. See second opinion for the full picture.
One source per review: screenshots, a live URL (desktop + mobile captures), a PDF deck, or email HTML. See how it works for the full loop.
Poll get_review and follow next_action: wait while you mark, fix your pins and spin up the next pass, or stop when you approve.
The board is your checklist: each mark moves open → resolved → verified. When the agent resolves marks, the review shows Awaiting your verify so you can check before/after and verify (or reopen). Request changes and your agent uploads fresh captures for pass 2 (and on) — the pass ledger keeps that history scannable. See the board.
Yep. Your review link is secret — anyone with it can mark and decide. Need a teammate or client? Use Share to copy or regenerate a guest link (suggestions only; your marks still run the show). Set expiry to 7 days (default), 14 days, never, or a custom date. See guest links.
06 Maker
I’m Derek — I love giving design feedback. Not to be a dick, but to be a Derek. Agents are getting fast at shipping UI; what’s still missing is a clear place for a human to mark what matters and send the next pass back without leaving the chat.
ReviseMy started as a side project in 2024 — an idea and a Figma file — then Taylor’s Laravel Cloud weekend challenge was the nudge to ship it as an MCP on Laravel. Built in a weekend on the side; it works, it passes tests, and there’s plenty left to improve. Sponsor on GitHub
Source is on GitHub under the O’Saasy License — use it, fork it, file issues, send PRs.
Say hi anytime — derekj@hey.com or @heyderekj on X.com.