Show your AI
what you mean.
Stop describing what's on your screen. Point at it, add one line, and Claude Code, Cursor or any MCP agent gets the actual picture — then fixes it and tells you. Your team works from the same board.
Point at it
One shortcut. Drag over the thing — a bug, a layout that's off, a design you want matched. Type one line. Peekr keeps the app, window and URL for you. No screenshot to take? Press it twice and just write.
Your agent sees the picture
Claude Code, Cursor, VS Code, Windsurf — any MCP client gets the screenshot and your markup, not a paragraph describing it. It claims the item, fixes it, and resolves it with the PR, attributed to you.
Your team works from the same board
Designers, PMs and clients point at what's wrong; developers and their agents pick it up. Groups control who sees what — guests see only theirs. Assign to a person, or to a bot.
The fastest way to get what you see into your agent's hands.
A screenshot pasted into a chat is gone in an hour. A Peekr item is a task with the image attached: every MCP client gets the same tools — list what's open, look at it, claim it, comment with a diagnosis, resolve it with the commit. On the board it reads Claude Code · Ada, so nobody wonders who did what.
- Hosted MCP with OAuth — one command, sign in once, scoped to what you can see.
- A local MCP server in the desktop app, with no network on the request path.
- Bots with webhooks, so an item can become a PR while you're at lunch.
$ claude mcp add --transport http peekr https://mcp.peekr.dev/mcp $ claude > what's open in the checkout group? ● 2 open captures in Checkout: #a1c3 “Wrong total after removing a coupon” — open, unassigned #9f02 “Apple Pay sheet shows twice” — open, assigned to you > look at the first one and fix it ● Looking at the screenshot… the arrow points at the total after the coupon line is removed: the discount is subtracted twice. Claimed #a1c3. Fixing cart/totals.ts…
Anything you'd otherwise have to explain.
Bugs are the obvious one. The habit that sticks is pointing at anything on screen and letting the picture do the talking.
Fix the UI you're looking at
"The save button is clipped at 125% zoom." Drag over it. The agent sees the clipping, finds the CSS, opens the PR.
Make it look like this
Screenshot a design, a competitor, a mockup. "Match this spacing on the settings page." Reference and instruction in one item.
QA rounds before launch
Walk the product, capture everything that's off, file it into a group. Hand the group to a person — or to a bot that works through it overnight.
Client and stakeholder feedback
Invite a client as a guest into one private group. They point; your team and agents fix; they see it resolved. Nothing else is visible to them.
Built like a tool you'd keep.
Native on Mac and Windows, zero dependencies, fast. The cloud part exists to share and to let agents in — not to get between you and your screenshots.
Agents receive the image with your arrows, boxes and blurs burned in. What you saw is exactly what they work from.
Nothing is shared by accident. New groups are private; guests are walled off; the database enforces the rule, not the UI.
Field-level merge across devices. When two people edit the same note offline, both versions land on the capture.
A bot has an identity, a role and a webhook. Make it a group's default assignee and the group becomes an agent's queue.
Signed events, retries for a day, scoped tokens. Wire captures into CI, Slack, or whatever you already run.
Native apps with no telemetry. Delete a capture and the bytes go. Export any time. Free while we're in early access.
Free while we're in early access.
Every feature, every teammate, every agent. When paid plans arrive you'll hear about it well in advance, and what you've captured stays yours and exportable either way.