06-ai-integration

Module 06 — AI Integration (optional, ~10 min)

Use AI tooling to interact with your local AWS environment — no cloud credentials needed.

---

Option A — LocalStack MCP Server

The LocalStack MCP server exposes your local AWS resources to AI assistants (Claude, Cursor, etc.) via the Model Context Protocol.

Setup (Claude Code / Claude Desktop)

Add to your MCP config (~/.claude/claude_desktop_config.json or .mcp.json):

{
  "mcpServers": {
    "localstack": {
      "command": "localstack",
      "args": ["mcp", "start"]
    }
  }
}

What You Can Ask

Once connected, ask Claude natural-language questions about your local stack:

"List all Lambda functions and their last invocation status"
"Show me the messages in the orders-dlq queue"
"Scan the orders DynamoDB table and summarize the order statuses"
"What's in the order-receipts S3 bucket?"

---

Option B — LocalStack Skills (Claude Code)

Claude Code ships with built-in LocalStack skills. With LocalStack running, try:

/localstack          # manage LocalStack lifecycle
/localstack-logs     # analyze logs and debug errors
/localstack-state    # save/load state with Cloud Pods
/localstack-iam      # analyze IAM policies

Example: Debug a Lambda Error

1. Trigger a Lambda failure (e.g., send a malformed order payload)
2. Run /localstack-logs in Claude Code
3. Claude will fetch the Lambda logs and suggest a fix

---

Option C — Claude Code + App Inspector

Combine App Inspector traces with Claude Code for AI-assisted debugging:

1. Find a failed trace in App Inspector
2. Copy the trace ID
3. Ask Claude:

   "Here's a failed trace from App Inspector: <trace-id>
    The order-processor Lambda failed. Analyze the logs and suggest what went wrong."
   

---

Why This Matters

• No cloud credentials exposure to AI tools — everything stays local
• AI can read real state, not mocked data
• Fast iteration: ask → inspect → fix → redeploy in seconds locally