use case · claude
The better the context, the better Claude's answer, and nobody wants to type three paragraphs of it. Say it instead. When it is about something on your screen, send the annotated thing itself — or, with Pro, let Claude ask you for it.
Wherever Claude takes text, dictation works the same way: put the cursor in the message box, press Ctrl ` , talk, and your words appear as editable text — read it over, trim it, send it. A spoken minute is roughly three typed ones.
Thinking out loud rarely comes out tidy. The Prompt voice mode reshapes a spoken ramble into a clear request before it lands. It runs free on the built-in local model or your own Ollama server; Pro can route it through your own Claude account, so the model that tidies the prompt is the one that answers it.
A design draft that needs feedback, a spreadsheet cell that makes no sense, a clause in a contract, a diff in a code review. Press Alt ` , drag over the region, draw a box around the part in question while you ask about it — and paste. Claude.ai and Claude Desktop receive the annotated screenshot and your transcribed words together, in one go.
In the terminal, delivery adapts: Claude Code gets your words plus the screenshot's file location — which is exactly what a terminal can use, and Claude Code reads the image from there. How delivery adapts per app →
Turn on the agent bridge and CroppickVoice becomes a local MCP server — MCP, the Model Context Protocol, is the open standard that lets AI apps use tools offered by programs on your machine. Claude Code and Claude Desktop connect with one click in Settings. From then on Claude can read your latest capture, ask you for a snip (your normal overlay opens and you choose what it sees), receive captures as you take them if you allow it, and ask you a quick question you answer by speaking.
The bridge answers only on your own computer, is off by default, and every ability has its own switch. One thing to know: whatever Claude reads through it becomes part of your conversation with Claude, which runs in the cloud. The setup guide →
Yes. Dictation types at the cursor in any app, terminals included: put the cursor on the prompt line, press the shortcut, and speak the instruction. Crop & Speak hands a terminal your words plus the screenshot's file path, which Claude Code can open.
No — the bridge is part of Pro, a one-time €39 purchase. Dictating into Claude and Crop & Speak are free, with no word caps and no account.
No. Speech becomes text on your device. Claude sees exactly the text you send or the screenshot you paste — and, with the bridge on, only what you have switched on and confirmed.
Yes. Dictation works in any editor the way it works anywhere else, and the agent bridge accepts any MCP-capable app — Cursor and VS Code get one-click connectors next to Claude's.
Dictation and Crop & Speak are free; the bridge rides along when you go Pro. Related: dictating to ChatGPT, asking AI about your screen, connecting your AI apps and all use cases.