How do you give an AI agent the ability to extract tasks from text (MCP)?
Install an MCP server that exposes extraction as a tool. actions-xyz-mcp gives Claude, Cursor or any MCP client a single tool — extract_action_items — that turns transcripts or notes into structured tasks: npx -y github:Blockchainpreneur/actions-xyz-mcp. No API key required.
Setup in any MCP client
Add the server to your client’s MCP config with command "npx" and args ["-y", "github:Blockchainpreneur/actions-xyz-mcp"]. The tool accepts a transcript string and returns JSON action items plus a readable summary. Rate limits: 10 calls/day per IP on the free endpoint.
Why agents need structured extraction
Agents that summarize meetings produce prose nobody executes. Structured extraction (owner, priority, deadline, steps) lets the agent file tasks, draft follow-ups, or populate boards — output another system can act on. This server is also dogfood: the product it belongs to is itself operated by an autonomous agent (see the build log).
FAQ
Does it work with Claude Code?
Yes — add it via the standard MCP config; the README shows the exact JSON for Claude, Cursor and Windsurf.
Is it open source?
Yes, MIT: github.com/Blockchainpreneur/actions-xyz-mcp.