Get started with Klickbrain.
Install the desktop app, add the browser extension, point your AI at the local MCP server, and record your first skill.
- 01
Install the desktop app
The desktop app runs locally, stores your recorded skills in SQLite, and exposes them as MCP tools on http://127.0.0.1:8731. Pick the build that matches your OS.
.deb (Debian / Ubuntu).rpm (Fedora / RHEL)macOS — Apple SiliconmacOS — IntelWindows installer (.exe)Current version: v0.8.3 · Free · Local-first.
- 02
Install the Chrome extension
The Klickbrain Recorder extension captures click, input, and form events while you teach a skill, and replays them back into the page on demand. It talks only to your local desktop app — no external servers.
Add to ChromeTip: pin the extension to your toolbar after install — click the puzzle icon in Chrome → pin Klickbrain Recorder.
- 03
Pair the two
Launch the desktop app. Click the Klickbrain icon in your browser toolbar — the popup should show a green Connected dot within a second. The extension and desktop app communicate over a fixed loopback WebSocket (127.0.0.1:8731). Nothing leaves your machine.
- 04
Add your AI API key (optional but recommended)
Klickbrain replays skills deterministically first. When a recorded selector no longer matches the page (the DOM moved, an element renamed), an optional Claude-powered fallback resolves the step. Add your Anthropic key in desktop Settings → AI.
Where to get a key: console.anthropic.com/settings/keys — sign in, click Create Key, copy it once.
Where it lives: Encrypted in your OS keychain (DPAPI on Windows, Keychain on macOS, libsecret on Linux) via Electron
safeStorage. Klickbrain never sees or transmits it. Details in our privacy policy. - 05
Connect your AI client over MCP
Register the local MCP server with your AI client. Every saved skill becomes a callable tool by name.
Claude Code
claude mcp add --transport http --scope user klickbrain http://127.0.0.1:8731/mcpCursor
Add to ~/.cursor/mcp.json: { "mcpServers": { "klickbrain": { "url": "http://127.0.0.1:8731/mcp" } } }Codex
Add to ~/.codex/config.toml: [mcp_servers.klickbrain] url = "http://127.0.0.1:8731/mcp"Restart your AI client after registering so it discovers the new tools.
- 06
Record your first skill
Pick something tedious you do every week. A good first skill: search YouTube for a phrase and save the top result to Watch Later.
- Open the page where the skill starts (e.g.
youtube.com). - Click the Klickbrain icon → press Start recording.
- Click through the flow exactly as you would normally.
- Press Stop. The desktop app opens with the captured steps.
- Give the skill a name (
youtube_save_to_watch_later), a one-line description, and add any parameters you want to vary (like the search query). Save.
- Open the page where the skill starts (e.g.
- 07
Replay it from your AI
In Claude Code, Cursor, or any MCP client, just describe what you want: "save the latest Rich Hickey talk to Watch Later". The AI invokes your skill by name, passes the parameters, and the browser executes the recorded steps.
That's the loop. From here you can record any flow — daily standups, expense imports, GitHub triage, scraping dashboards. Every skill you save is one prompt away from running.
Troubleshooting
›Extension popup shows Disconnected
The desktop app isn't running. Launch it from your applications menu or system tray. The extension reconnects automatically within a few seconds.
›Replay failed mid-skill
A selector no longer matches the page. Open Settings → AI in the desktop app and confirm your Anthropic key is saved; that enables AI fallback on the next run. Alternatively, re-record the affected step.
›Skill not appearing in Claude / Cursor
Restart the AI client after running the MCP add command. Then run claude mcp list (or your client's equivalent) to confirm klickbrain is listed.
›Port 8731 already in use
Another Klickbrain instance is running, or another tool grabbed the port. Quit the other instance or, if needed, kill the process on 8731 and relaunch the desktop app.
›Still stuck?
Email team@klickbrain.com with what you tried and the OS you're on — we'll get back to you within a day.