From zero to AI-driven board.

1. Create your workspace

Head to letsweft.com/sign-up and finish the form. You will land on an empty board with four columns already provisioned: Backlog, Todo, Doing, Done.

Free accounts get 50 tasks on the board and 200 AI calls a month (the first 48 hours are unmetered, so an import lands in full) — plenty for the quickstart and a few sprints. No credit card.

2. Connect your AI

Weft exposes a single MCP endpoint, https://letsweft.com/api/mcp, protected by OAuth 2.1 with Dynamic Client Registration. That means your AI client can discover and register itself — you just sign in once in a browser popup.

Pick your client from the integrations page and copy its configuration snippet. A few common ones:

• Claude Code — one shell command: claude mcp add --transport http weft https://letsweft.com/api/mcp
• Cursor — one-click “Add to Cursor” button on the setup page, or a JSON entry in ~/.cursor/mcp.json.
• ChatGPT — Weft is a published plugin: Plugins → search “Weft” → Connect. Then @-mention it: “@Weft show my board”.

3. Ask your AI to do something

Open your AI client and try any of these prompts. The Model Context Protocol treats tools as model-controlled: a server exposes executable functions and the model decides when to invoke them, which is why a plain sentence is all you need to type. Your AI will call the right Weft tool, and you will see the result on the board in real time.

• “What's on my Weft board?”
• “Add a high-priority task: write the launch post.”
• “Move the auth refactor into Doing.”
• “Start a one-week sprint focused on shipping the landing.”
• “Complete the current sprint and archive everything.”

The first three will land instantly; the last two affect sprint state and your AI should confirm before calling them.

What to read next

• Scrumban — the method Weft is shaped around.
• MCP tool reference — every tool your AI can call, with arguments and examples.
• Built for — how solo founders, indie hackers, and small teams run Weft day to day.