본문으로 건너뛰기

Integration Assistant Overview

The Integration Assistant is a powerful feature in the FluxLoop VSCode extension that helps you integrate FluxLoop SDK into your project with AI-guided suggestions.

What is the Integration Assistant?​

Integration Assistant combines four key technologies:

  1. FluxLoop MCP Server – Provides documentation, recipes, and framework-specific workflows
  2. Repository Analysis – Runs run_integration_workflow to detect languages, frameworks, entry points, and risk flags
  3. Mode-Specific Context APIs – Integration, Base Input, Experiment, and Insight payloads (Integration mode available today; others in preview)
  4. LLM Suggestions – Uses OpenAI Chat Completions models to generate contextualized integration plans

Key Features​

🔍 Knowledge Search​

  • Ask documentation questions directly from VSCode
  • Answers include file citations and section names
  • Great for quick “how do I configure X?” lookups before running the agent

🤖 Flux Agent​

  • Analyzes the folders/files you select (no need to open the file)
  • Combines MCP repository data + mode-specific context with OpenAI reasoning
  • Generates structured plans (summary, suggested edits, validation checklist, references)
  • Keeps the human-in-the-loop: suggestions are review-only

📊 System Status​

  • Real-time MCP connection and knowledge index checks
  • Environment + dependency validation (CLI, SDK, MCP)
  • One-click access to FluxLoop: Run Doctor

📝 Suggestion History​

  • Stores your last five suggestions per project
  • Reopen any entry with full context and copy buttons
  • Helpful for comparing iterations or sharing plans with teammates

How It Works​

graph TD
    A[Select Files/Folders] --> B[Run Flux Agent]
    B --> C[Collect Project Context]
    C --> D[MCP Analysis + Mode Payload]
    D --> E[LLM Generation]
    E --> F[Render Plan + History]
    F --> G[User Reviews & Applies]
  1. Context Collection – You pick folders/files; the agent also reads source_root, project metadata, and selection snippets (if any).
  2. MCP Analysis – Runs run_integration_workflow plus mode-specific context tools (Integration today; Base Input/Experiment/Insight in preview) to gather repo profile, detection results, and suggested edits.
  3. LLM Generation – Sends all context to the configured OpenAI Chat Completions model to craft a plan.
  4. Presentation – Renders Markdown with summary, suggested edits, validation checklist, and citations. Copy buttons make applying changes easy.
  5. Manual Application – You review, copy, and apply changes manually in your codebase.

Benefits​

  • Faster Integration: Get framework-specific guidance in seconds
  • Accurate Recommendations: Based on your actual code structure
  • Safe: No automatic modifications, you control all changes
  • Learning Tool: Understand why and where to make changes
  • Traceable: Full citation chain from suggestion to source docs

Prerequisites​

Before using Integration Assistant, ensure you have:

  • Python 3.11+ (Flux Agent uses the MCP server)
  • fluxloop-cli, fluxloop, and fluxloop-mcp installed in the active environment
  • MCP knowledge index built (packages/mcp/scripts/rebuild_index.sh)
  • configs/project.yaml includes the correct source_root (set via Target Source Root…)
  • OpenAI API key (prompted on first use; securely stored if you allow it)

Next Steps​