Product Launch Agent — single-tool checkpoint

This is the small application built in the course around slides 19–26, and the first interim step toward the completed Product Launch Agent.

It does three important things:

1. Sends a launch brief to Claude.
2. Gives Claude one available tool, find_launch_assets.
3. Runs an agent loop: when Claude requests the tool, JavaScript executes it, sends the result back to Claude, and lets Claude decide what to do next.

At this checkpoint the app does not yet generate release notes, blogs, social posts, outreach, image crops, saved run results, or the launch-kit report. Those capabilities are added in the next course videos, building up to the completed project linked above.

Files

• src/index.js — starts the demo with a launch brief.
• src/agent.js — calls Claude, executes requested tools, returns results, and repeats until Claude gives a final answer.
• src/tools/tool-definitions.js — defines the one tool Claude can see and routes requests to JavaScript.
• src/tools/cloudinary-tools.js — implements the real Cloudinary search operation.
• src/config.js — configures Anthropic and Cloudinary clients.

Run it

npm install
cp .env.example .env
# Add your credentials to .env
npm start

Or provide your own brief:

npm start -- "Find the assets for our summer-accessories-2026 launch and summarize what you found."

What to notice

The key distinction is that Claude chooses the tool, but JavaScript executes it. agent.js repeats that round trip until Claude no longer asks for a tool. That repeating round trip is the agent loop.