Best Practices & Prompting Guide

FlowLens MCP is designed to give AI coding agents the right context from your recorded flows—delivering only what’s needed without overloading the context window. Once connected, you can ask them to investigate, debug, analyze, find the root cause, etc., all in natural language—there are no rigid rules or templates to follow.

How to Start

Step 1: Click “Debug with AI”

In the flow detail view, click the “Debug with AI” button
You’ll see two options:
- Copy Flow ID: Just the flow ID for maximum flexibility with your prompts
- Copy Debugging Prompt: Flow ID + pre-written prompt that finds the 3 most likely root causes

Step 2: Prompt Your AI Agent

Open your AI coding agent (Claude Code, Cursor, GitHub Copilot, etc.)
Make sure you are in the root folder of the relevant coding project
Include the flow ID in your prompt and let the agent do the debugging

Common prompt examples:

“Investigate what is wrong with flow copied-flow-id”
“Find the root cause of the issue in flow copied-flow-id”
“Fix the bug highlighted in flow copied-flow-id”

What’s available via FlowLens MCP: Flow metadata (title, description, comments), technical timeline (network requests/responses, console logs, errors, user interactions, DOM actions, storage changes), visuals (screenshots at any timestamp).

Tips for Using FlowLens MCP

Give your flow a clear title and description - This helps the AI understand the context
Mention expected behavior in your flow description or add comments at specific timestamps to highlight what should have happened
Stay in your project’s root folder when prompting so the AI can easily correlate the recorded flow with your codebase

Share Your Discoveries

Have you found useful prompts or debugging techniques with FlowLens MCP? Share them with us at flowlens@magentic.ai—we’ll add the best ones to help the community!