setup guide · mcp
Connect your AI apps.
CroppickVoice can act as a local MCP server — the open standard AI apps use to reach tools on your computer. Connect any agent that speaks MCP — Claude Code, Claude Desktop, Cursor, VS Code, or anything else — and it gets eyes on your screen and a voice line to you: it sees what you capture, can ask you for a snip, and can ask you a question you answer by speaking. On this computer only, gated by switches you control. A Pro feature.
what this unlocks
Once connected, you just talk to your AI app the way you already do. Try: “Grab my last CroppickVoice capture” · “Snip the error on my screen for me” · “Ask me by voice which option I prefer.”
- See what you capture. The agent reads your latest Crop & Speak capture — the picture plus what you said over it — and can look back through your dictation history and earlier captures.
- Ask for a look. The agent asks; your normal snip overlay opens. You choose the area, redact anything private, confirm — or press Esc to say no.
- Get captures as you take them. Switch on capture sharing and every snip you take lands in the agent's session by itself — no copy-paste, no “here's a screenshot”.
- Take an aimed capture. If you allow it, the agent can capture on its own — a window by name, the area you last snipped, or an exact spot on screen — every shot announced with a flash and a notification.
- Ask you a question. A short question appears beside the floating button; you answer by speaking and only the agent hears it. Quick decisions, no keyboard.
- Always in the open. Nothing is shared silently: every capture and answer is marked in History, you can decline any request, and each ability has its own switch.
step 1 · turn the bridge on
- Right-click the floating button → Settings → Connect AI apps (MCP) (named “Agent bridge (MCP)” in v1.2).
- Enable the agent bridge and Save — the status line flips to ● Running.
- Leave CroppickVoice running — connected apps can only reach it while it's up.
Shortcut: Quick setup for a coding agent on the same page — one confirmation turns on the bridge and every ability, so an agent like Claude Code has everything at once.
step 2 · one click per app
The same page lists the AI apps found on your computer, each with a Connect button. Click it, and watch the line under the row — it flips to ✓ Connected the first time that app actually reaches the bridge.
- Claude Code — Connect registers the bridge with the
claudeCLI for you (no terminal). Open a new Claude Code session;/mcpshould listcroppickvoice. - Claude Desktop — Connect opens a small connector file in Claude Desktop; approve its install review. If the tools don't show up right away, restart Claude Desktop once.
- Cursor / VS Code — Connect opens the editor's own “install MCP server” prompt; approve it there.
Several apps at once is fine — every app you connect can use the bridge at the same time.
any other mcp app
Open Setup steps & other apps on the same Settings page. It shows both
standard forms, prefilled with your real install paths: a stdio command (the
app's own executable run with --mcp — no token to copy, it finds the running app
by itself) and a Streamable HTTP endpoint (http://127.0.0.1:27182/mcp with a bearer token from the same page). Any
MCP-capable client works with one of the two.
if something doesn't connect
- “CroppickVoice isn't running.” That's the honest answer your AI app gets when the bridge is down — start CroppickVoice and ask again.
- A tool is missing from the list. Its switch is off on the Settings page — each ability is opt-out individually.
- Claude Desktop lists the server but calls fail. Restart Claude Desktop once after installing the connector.
- You regenerated the access token. Only hand-pasted HTTP configs break — update the header. One-click connectors pick the new token up on their own.
- The page is greyed out. The bridge is part of Pro — one payment, yours forever.
private by design
The bridge answers on 127.0.0.1 only and never connects out; every request needs
its access token. An AI app can ask for a screenshot, but you always take it — unless
you deliberately switch on the separate options that let it receive the captures you take or
take aimed captures itself, each announced with a flash and a notification. A question from an
agent is only answered when you press Answer and speak. One thing to know: whatever a connected
app reads becomes part of your conversation with that app's AI model, which usually
runs in the cloud. The full network-use table →
Not set up yet? Download CroppickVoice — dictation and Crop & Speak are free forever; the bridge rides along when you go Pro.