Client website feedback, straight into Cursor.
If you run an agency, the revision round is where margin goes to die: the client sends an email with eleven numbered points, four screenshots, and "the thing near the top feels off." Someone on your team spends an afternoon translating that into work. Here's the version where the client's click is the work item, and your coding agent implements it.
Anatomy of a revision round, today
- You send the staging link. The client looks at it on their phone, in their inbox, between meetings.
- Feedback arrives across three channels: an email, a Loom, two WhatsApp messages to your PM.
- Your PM consolidates it into a doc. Some items are ambiguous, so a clarification call gets scheduled.
- A developer reads the doc, opens the repo, hunts for each element, and makes the changes — or increasingly, re-describes each item to Cursor or Claude Code and pastes in the client's screenshots.
- You reply "all done!" and the client responds with the items that got lost in translation. Round two.
Notice the shape: the client already pointed at the exact element in their browser, and by step 4 that precision is gone — reconstructed by hand so an AI agent can guess at it. The middle of this pipeline is pure translation loss.
The client clicked the exact element. Your agent needs the exact element. Everything between those two facts is overhead.
The pin-based version
1. The client pins the live site — no installs, no accounts to buy
Send the same staging or production link, plus either the Chrome extension or nothing at all — if you've added the one-line script-tag widget to the site, anyone who opens it can pin. The client clicks the element, types the note, done. Clients and reviewers are free and unlimited on every plan; they never consume a seat. On Pro you can lock commenting to the client's email domain or an invite list, so a forwarded link doesn't become a public suggestion box.
2. Each pin is already a work packet
The pin captures the CSS selector, a DOM snippet, a screenshot, the viewport it was seen at, and the thread. "The thing near the top feels off" becomes header .hero-cta at 390×844 with a screenshot attached. Ambiguity drops to near zero before anyone on your team reads anything. Discussion happens on the pin, so clarifications don't spawn email threads.
3. You review, approve, and run /implement
Approval is your quality gate — nothing a client pins auto-ships. Then, in Cursor or Claude Code:
/implementThe agent calls implement_approved_pins over MCP, gets the pins grouped by page (one page = one branch = one PR), greps the codebase using each selector, makes the changes, and calls fix_and_resolve with the commit SHA, branch, and PR URL. A ten-item revision round is one agent run plus your code review of the diff.
4. The client sees receipts, not promises
When the deploy goes out, the deploy hook links the live URL to each resolved pin. You can also mint a shareable crit report — a public read-only link showing every pin with its thread, status, and the branch/PR/deploy trail. Send it with the invoice. "All eleven items shipped, here's the deploy for each" is a different client conversation than "all done!"
What changes for the agency, concretely
| Email + screenshots | Pins + agent | |
|---|---|---|
| Consolidating feedback | PM afternoon per round | Zero — pins arrive structured |
| Ambiguity / clarification calls | Routine | Rare — selector + screenshot + viewport attached |
| Implementation | Dev re-describes each item to the agent | One /implement run, then code review |
| Client-facing status | "We're on it" | Live report: pin → PR → deploy, per item |
| Items lost between rounds | The classic failure | Pins stay open until resolved — nothing evaporates |
Where the ticket tools fit in (and where they don't)
Client-feedback tools like Pastel, Markup.io, and Feedbucket solved step 2 — they get client comments out of email and into one organized place, often with an export into Trello or Asana. That was the right target when a human developer was going to read each item. If your delivery runs through Cursor or Claude Code, the dashboard is a layover: the feedback still has to be re-translated for the agent at the end. Pincushion's bet is that the feedback should be captured in the agent's format from the start — that's the difference between a comment tool with integrations and an MCP server with a feedback UI.
Cost, for an agency
Clients never pay and never eat a seat. The free tier covers one project with unlimited manual pins and 25 AI actions a month — enough to run the loop on a single client site and see if it sticks. Pro is $19/mo (14-day trial on signup) and adds unlimited projects — one per client — plus realtime push to your IDE, access controls (domain-restricted or invite-only commenting), and post-deploy verification. That's less than the hourly cost of one consolidation afternoon, per month.