Skip to main content
AI can build your AI. Whether you’re shaping a feature or tuning a prompt, your assistant can create prompts, workflows and evaluations in PromptJuggler, run them, and improve them with you. It connects through MCP, the open standard for plugging AI assistants into the tools you use. Connect it once, then ask in your own words:
  • “Draft a prompt that sorts support tickets by urgency, try it on these three tickets, and show me the results.”
  • “Compare v3 and v4 of the summariser on these inputs and tell me which handles the edge cases better.”
  • “Build a workflow that drafts a reply and has a second prompt check its tone before returning it.”
  • “Find yesterday’s failed runs of the triage workflow and tell me what went wrong.”
  • “What can you do in PromptJuggler?”

​
Connect over MCP

PromptJuggler’s MCP server has one address for every assistant: https://mcp.promptjuggler.com. Working in more than one workspace? Each has its own address, which Connected apps on My Account lists for you to copy. Add one connection per workspace with its address, and name each connection after its workspace.
On claude.ai or in the Claude desktop app:
  1. Go to Customize > Connectors.
  2. Click Add, then Add custom connector.
  3. Name it PromptJuggler, enter https://mcp.promptjuggler.com as the server URL, and click Continue.
  4. Keep the settings Claude detects, and click Add.
  5. Click Connect.
On a Team or Enterprise plan, an Owner adds the connector once under Organization settings > Connectors. You then find it under Customize > Connectors and click Connect.Sign in, pick the workspace, and choose what your assistant may do.

​
You stay in control

  • You choose what it can reach. Each connection works in the one workspace you pick, and you decide whether your assistant may edit drafts, start runs and publish.
  • It asks before publishing. Publishing can change what your product runs, so your assistant checks with you first.
  • You can disconnect it any time. Connected apps on My Account lists your assistants and tokens, so you can disconnect an assistant or revoke a token whenever you like.

​
From idea to product

Your assistant works with the same drafts, versions and tags you do. Naming & Versioning explains them; here’s how they take an idea all the way to your product. Publish once it works. Drafts are yours alone, so you and your assistant can iterate freely. Publishing makes a version your team and your product can use. Choose how your product picks up changes. Your product can call a version number, so changes ship with your own releases. Or it can call a tag such as production, so changes ship, and roll back, without a deploy. Both work well. A common setup pairs a staging and a production tag with matching environments, so each gets its own variables and webhooks. Wire it in. Ask Claude Code or Codex to call it from your code, or ask your assistant for a handover note for your developers. Every step is one request away:
Publish this as “shorter answers” and run the ticket evaluation on it and on the version in production.