Building a Tech-Radar Agent with GitHub Trending MCP

What this pipeline does

This guide turns the HarvestLab GitHub Trending MCP actor into a weekly technical radar. The actor collects trending repositories, normalizes repository metadata, and exposes a compact schema that an LLM can rank without reading GitHub HTML.

The pipeline is simple: run the actor, keep the fields that explain velocity, cluster repositories by language or topic, then ask your agent to produce a short memo with citations back to the original repository URLs.

Output fields to keep

{
  "repo": "owner/name",
  "url": "https://github.com/owner/name",
  "description": "Short repository description",
  "language": "TypeScript",
  "stars": 18420,
  "stars_today": 612,
  "topic_cluster": "agent tooling"
}

Agent prompt pattern

Use a ranking prompt that rewards velocity and penalizes generic projects. The agent should return a structured list with the repository name, why it matters, who should investigate it, and the source URL.

const radarPrompt = `
You are building a weekly technical radar for AI engineers.
Rank repositories by practical agent-builder relevance.
Return JSON with repo, trend_reason, ideal_user, and source_url.
`;

Deployment notes

Run the actor on a daily or weekly schedule, store each dataset run, and compare stars_today against prior runs before the agent summarizes. This keeps the memo focused on new movement instead of large projects that are always popular.