惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Y
Y Combinator Blog
Google DeepMind News
Google DeepMind News
腾讯CDC
V
Visual Studio Blog
Engineering at Meta
Engineering at Meta
博客园 - 司徒正美
小众软件
小众软件
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
T
Tailwind CSS Blog
Vercel News
Vercel News
爱范儿
爱范儿
Last Week in AI
Last Week in AI
G
Google Developers Blog
阮一峰的网络日志
阮一峰的网络日志
P
Proofpoint News Feed
有赞技术团队
有赞技术团队
D
DataBreaches.Net
博客园_首页
J
Java Code Geeks
云风的 BLOG
云风的 BLOG
V
V2EX
A
About on SuperTechFans
H
Hackread – Cybersecurity News, Data Breaches, AI and More
人人都是产品经理
人人都是产品经理

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
A non-engineer CEO tried to connect Copilot Studio to LDX...
Kozo-KI · 2026-05-07 · via DEV Community

I'm the CEO of a translation company. I'm not an engineer. I can read code, but I can't write it.

That's the context for everything that follows.

LDX hub recently launched plugins for n8n, Dify, and Copilot Studio, making it possible to call StructFlow — their structured data extraction API — from no-code workflow tools. I decided to try it myself instead of delegating.

Spoiler: it's not working yet. But I'm publishing this anyway, because a record of where things broke is more useful than a "it all went smoothly" writeup.


What I was trying to build

Goal: An agent in Copilot Studio that takes meeting minutes as input and returns structured action items.

Stack:

  • Microsoft Copilot Studio (agent)
  • Power Automate (flow / middleware)
  • LDX hub StructFlow (REST API)

Approach: REST API first. If that works, compare with MCP.


The architecture

User (Copilot Studio chat)
    ↓  pastes meeting minutes
Power Automate flow
    ↓  POST to StructFlow API
LDX hub StructFlow
    ↓  returns structured JSON
Power Automate flow
    ↓  passes result back
User (receives action items)

Enter fullscreen mode Exit fullscreen mode

Straightforward on paper. Less so in practice.


Building the Power Automate flow

Finding the right trigger

Opened make.powerautomate.com, selected "Instant cloud flow."

First blocker: "Power Virtual Agents" trigger wasn't visible.

There was no search box on the trigger selection screen. I scrolled through the list and eventually found it. The label was slightly different from what the documentation described. Found it, selected it, moved on.

Input variable

Added a text input variable called minutes_text to the trigger. No issues here.

HTTP action

Added "Send an HTTP request" and configured:

  • URI: https://your-ldxhub-host.d2.zuplo.dev/structflow/jobs
  • Method: POST
  • Header: Authorization: Bearer {API_KEY}

Second blocker: the header field format.

Instead of separate "key" and "value" fields, this action uses a single text field per header, formatted as HeaderName: HeaderValue. So the Authorization header goes in as:

Authorization: Bearer your-api-key-here

Enter fullscreen mode Exit fullscreen mode

Not obvious if you're used to other HTTP tools.

Polling loop

StructFlow is asynchronous. You POST a job, get back a job_id, and need to poll until status: "completed" before you can retrieve results.

In Power Automate, this means a Do until loop. The structure:

  1. Initialize variable job_id (string)
  2. Set variable — extract job_id from HTTP response
  3. Delay 5 seconds
  4. Initialize variable job_status (string)
  5. Do until loop:
    • GET request to check job status
    • Set job_status from response
    • Stop condition: job_status equals completed
  6. Return result to Copilot Studio

Third blocker: the browser back button deletes your flow.

I hit the back button before saving. The flow was gone. Started over.

Lesson: save constantly in Power Automate. There's no autosave equivalent to what you'd expect from a modern web app.

The Do until condition also caused an error when I tried to reference job_status via the GUI. Fixed it by switching to "Edit in advanced mode" and typing the expression directly:

@equals(job_status, 'completed')

Enter fullscreen mode Exit fullscreen mode

The flow eventually completed. Green banner: "Your flow is ready to run."


Connecting to Copilot Studio

Created an agent called "Minutes Assistant" in Copilot Studio.

Observation: the default model was Claude Sonnet 4.6.

Our company uses Claude Team, so Microsoft apparently picked it up automatically. Our enterprise agent running on Claude — mildly amusing given the context of this article.

Now for the main blocker: the Power Automate flow didn't appear in Copilot Studio's tool list.

Same environment, same account. The flow just wasn't visible. I tried searching, browsing the Workflow tab — nothing.

Explored the sidebar and found "Flows" → "Agent flows" — a feature where you describe what you want in natural language and AI generates a flow. The AI generated something instantly, but the action it created pointed to an unrecognizable internal ID — certainly not the StructFlow API in any recognizable form.


Where things stand

Step Status
Power Automate flow ✅ Built
Do until polling loop ✅ Working
Copilot Studio connection ❌ Unresolved
Agent flow AI generation ⚠️ Generated, not verified

What tripped me up:

  1. No autosave — browser back = flow gone
  2. Single-field header formatKey: Value in one box, not two
  3. Flow not visible in Copilot Studio — same environment, still not showing
  4. AI-generated flow — unclear what API it's actually calling

What's next

  • Inspect the AI-generated agent flow to see what it's actually calling
  • Test whether StructFlow is being invoked correctly
  • If it works: run it against real meeting minutes
  • Part 2 will cover the full resolution (and MCP comparison)

I'll publish the follow-up when it actually works. This one ends here — unresolved, on purpose.


Kawamura International is a translation and localization company. We're documenting our AI process experiments as we go — StructFlow, RefineLoop, RenderOCR, and whatever comes next.