I kept losing the small stuff while working in Cursor. Halfway through an agent run I’d notice a flaky test or a config that needs redoing, and the choice was either derail the task or trust myself to remember it later. I usually didn’t.
So I wired my GTD board into Cursor over MCP. Now I say “capture that” and the agent drops it in my Inbox and carries on with what it was doing. When I want to know what I should be working on, I ask for my next actions and it reads them back.
It isn’t an extension, so it won’t show up in your Extensions view. It’s a remote MCP server, which means it’s two lines in mcp.json:
{ "mcpServers": { "gtdbrain": { "url": "https://mcp.gtdbrain.com/api/gtdbrain/v1/mcp" } } }
Sign-in happens over OAuth from inside Cursor: click the server’s “Needs login”, enter the code we email you, then click Allow. There’s no API key to paste anywhere.
The same board is on the web and on iOS and Android, so whatever I capture mid-task is waiting for me when I sit down to actually sort through it.
I make this, and it’s a paid product. You get 3 free MCP actions to try it before it asks for anything.
Guide: Connect Cursor to GTD Brain — capture tasks from your editor
Happy to answer questions about the MCP side. I’m also curious whether other people capture things mid-run or just let them go.