Docs
MCP server
ScreenMind ships an MCP (Model Context Protocol) server that exposes your screen history as tools any MCP-compatible assistant can call — Claude Desktop, Cursor, and VS Code (via Cline/Continue) — over stdio.
Setup
1
Install the MCP extra
Run
pip install "mcp[cli]" in your ScreenMind environment.2
Point your AI client at ScreenMind
Add the server to your client's MCP config (below), using your install path.
3
Restart the client
"screenmind" appears with its tools, ready to query your history.
Client config
{
"mcpServers": {
"screenmind": {
"command": "python",
"args": ["-m", "screenmind.mcp_server"]
}
}
}
"mcpServers": {
"screenmind": {
"command": "python",
"args": ["-m", "screenmind.mcp_server"]
}
}
}
Config location — Claude Desktop: %APPDATA%\Claude\claude_desktop_config.json (Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (macOS). Cursor: Settings → MCP. VS Code: .vscode/mcp.json.
Tools
search_screen— natural-language search across your history (semantic + keyword)get_recent_activity— the most recent N activitiesget_activity_by_time— activities for a specific date/time rangeget_daily_summary— the AI daily summary and standup notessearch_audio— search across meeting transcriptsget_screenshot— fetch a screenshot path by activity idcapture_now— trigger an instant screenshotget_stats— overall statistics about your screen history
Test it
npx @modelcontextprotocol/inspector python -m screenmind.mcp_server
Opens the MCP Inspector web UI so you can call each tool interactively before wiring up your assistant.
Tip: ask Claude “what was I working on at 3pm?” — it calls
get_activity_by_time under the hood and reads straight from your local database. Read-only, all local, no API keys.Explore more
What's insideInstall & runChat & SearchVoice & meetingsAgentsNothing leavesUnder the hood