Debriefer
Recipes

Iterate on a blueprint by talking to it

Refine questions and branches without leaving your AI assistant.

You shipped a blueprint yesterday. The first ten interviews came back. The third question is weak — respondents keep asking what you mean.

Don't open a GUI. Don't open an editor. Fix it by talking to Claude.

Load the blueprint as context

Claude pulls the blueprint into the conversation as an MCP resource. You can refer to nodes by number and Claude knows exactly what you mean.

Say this to Claude Desktop

Load blueprint bp_7f3a_abc as context. Show me the current questions.

Refine a question

Say this to Claude Desktop

Rewrite node 3 to probe for a specific moment of friction instead of a generic "what could be better". Keep the same modality.

Add a conditional branch

Say this to Claude Desktop

If the user mentions pricing in node 3, add a follow-up that asks what would have felt fair.

Validate before republishing

Say this to Claude Desktop

Check the graph for problems before I publish.

Republish

Say this to Claude Desktop

Publish this as a new version. Give me the new hosted link.

Why this is uniquely MCP

A drag-and-drop GUI asks you to click. A natural-language interface lets you describe intent. "Add a follow-up only if they mention pricing" is one sentence here and twelve clicks anywhere else.

The underlying tools — add_blueprint_node, replace_edges, validate_graph, publish_blueprint — are the same ones any API client calls. The difference is the interface.

On this page