Image: Point and Yap
Why it mattersWriting a careful bug description for a coding agent is where most of a session goes, so a tool that takes a sentence and a crop instead keeps the loop short.
Writing a careful bug description for a coding agent is where most of a session goes. Point and Yap, released by Marc Maurer on 18 October, cuts that step out. Press Control twice, rest the cursor on the thing that is wrong, say what should change, press Control twice again. Claude Code, Codex or Cursor pick the recording up over MCP and fix it.
The Mac app is free and MIT-licensed, and ships at version 0.1.4. It opened for downloads on 18 October 2026 and reached Product Hunt the next morning.
What the recording captures
A tap of Control twice starts the capture and shows a small pill on top of every window with the time, Esc and Stop. The cursor becomes the pointer. Rest it on an element and talk; click if that helps. A second tap of Control twice ends the capture and a soft tone plays. Esc throws the capture away.
Each recording becomes an item in a local queue, and the queue carries three things: what you said, the element under the cursor at each point and click (its role, label, window and app, read through the macOS Accessibility API), and a cropped screenshot. The recording pill is never in the screenshot. The clipboard is set to two lines that tell any agent what to call to pick the item up.
The MCP server
The app is also an MCP server over stdio, named pointandyap. On first run it offers to add itself to Claude Code at user scope, Codex in ~/.codex/config.toml and Cursor in ~/.cursor/mcp.json. Each config file is backed up first, and an existing MCP entry already called pointandyap is left alone.
Six tools make up the loop. get_open_items returns the open items oldest first and marks them in progress by the calling agent. mark_done closes an item with a one-line note; a done item is never handed out again. release_items returns items to the queue with a reason. list_items, list_sessions, get_session and get_frame give the raw recordings, their aligned timeline and the full-resolution screenshots. One file lock keeps two agents from taking the same item, and an item left in progress for more than two hours returns to the queue.
Where the recordings live and where they go
Speech is transcribed by Apple's on-device SpeechAnalyzer, a macOS 26 framework. Recordings, screenshots, the queue and the MCP server all live on the Mac, in ~/PointAndYap/sessions. The app itself only goes online to check pointandyap.com for updates.
Once an agent calls get_open_items, the words you said and the cropped screenshots travel to whatever model it is pointed at, like anything else shown to it. The README is explicit about this.
Describing a visual bug in text loses the thing you were pointing at. Pairing the element with a spoken note gives the agent the widget by its accessibility identifiers and the intent in your own words, so the fix stays in one exchange.
Source
The project page is at pointandyap.com and the code is at mfmaur/pointandyap on GitHub, under an MIT license.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
