MCP — connect Claude, Cursor or VS Code to your workspace

Point an MCP client at https://mcp.productcraft.co/mcp and the workspace's API becomes tools the model can call: read a waitlist, look up why a delivery bounced, send a transactional email, run a web extraction. Claude Desktop, Claude Code, Cursor and VS Code all connect with the URL alone — the client registers itself, opens a browser, and you approve the connection on a ProductCraft consent screen that names the client, the workspace and exactly what it may do. No API key ever lands in a config file.

A connection carries a policy document, not a coarse scope. Start from a preset — readonly, mail, support or full — and narrow it before you approve. Whatever you grant is intersected with your own permissions on that workspace and re-checked on every call, so a demotion or an offboarding takes effect immediately rather than at token expiry. Privilege-escalating actions — minting credentials, editing roles, handing out seats, emailing outside your verified domains — are in no preset and have to be ticked one at a time. Owners and admins can list and revoke every connection made against the workspace, including by people who have left.

Every tool call lands in the workspace audit log, refusals included, marked mcp_client so "you did this" and "an AI client acting as you did this" are different rows. Calls are rate-limited per connection and per workspace, results are capped and paged, untrusted content comes back labelled as data rather than instructions, and secrets are stripped from every result. Setup, the per-client config snippets, the scope table and how to revoke are in the MCP docs.