✎ PROWRIT for your AI agent

Your agent can use the studio too.

Prowrit is an MCP server: Claude Code, Cursor, or any MCP client can send it a brief mid-task and get back a complete, reviewed workflow — sectioned prompts, skill files, quality checks — that the agent can execute immediately or hand to you. And it engineers for your agent's actual environment: the tools it reports are the tools the workflow uses. Same Engineer as the studio, same account, same private workspace.

engineer_workflow(brief, environment?)

Starts the real engineering run and returns a job id. Your agent describes the task the way you would — and reports its environment (its MCP servers and tools, platform, relevant context). With it, the workflow binds to real capabilities: “use the gmail tool to fetch unread messages”, evidence-checked steps the agent runs itself, setup for anything missing. Without it, prompts stay copy-paste-ready for any chat.

get_workflow_result(job_id)

Progress while it runs (simple asks in seconds; a full design usually 3–7 minutes); when done — the compiled workflow as ready-to-run markdown, the assumptions the Engineer made, and a link to the visual blueprint. If something load-bearing is genuinely unclear, the Engineer asks first (up to 3 questions with likely answers) — your agent answers from context or asks you, then re-calls. The finished workflow also lands in your studio for later refinement.

1 · Create your MCP key

Sign in, open My studio → Account → MCP access, and create a key. It's shown once, tied to your account, and you can rotate or revoke it any time.

2 · Add Prowrit to your client

Claude Code — one command:

Any MCP client that speaks Streamable HTTP — JSON config:

3 · Use it mid-task

Tell your agent something like: “Use prowrit to engineer a workflow for turning our weekly support tickets into a trends report, then run it.” The agent starts the run, polls until the blueprint is ready, and receives prompts it can execute step by step.

Good to know

· A full design takes 3–7 minutes — your agent polls (~every 20s); that's by design, the Engineer is doing real work.
· The key is a bearer credential: keep it in your client config, not in shared repos.
· Everything your agent engineers is saved to your studio — refine it there whenever you want.