gh-trends

Fetch, analyze, and digest GitHub trending repositories — from the command line, as an MCP server, or via a public API.

Quick start

# Install
uv sync

# Fetch today's trending Python repos
uv run gh-trends fetch -l python -w daily

# AI-summarized digest (requires ANTHROPIC_API_KEY)
export ANTHROPIC_API_KEY=sk-ant-...
uv run gh-trends digest -l python -w weekly --save

CLI commands

Command	Description
gh-trends fetch	Fetch trending repos as a rich table
gh-trends digest	Claude-summarized thematic digest (ko/en)
gh-trends serve	Launch MCP server (stdio)
gh-trends serve -t streamable-http	Launch MCP server (HTTP on :8000)
gh-trends version	Print version

fetch options

-l, --language    Language slug (python, rust, go, ...). Omit for overall.
-w, --window      daily | weekly | monthly (default: daily)
-n, --limit       Number of repos to display (default: 15)

digest options

-l, --language    Language slug
-w, --window      daily | weekly | monthly (default: weekly)
--lang            Output language: ko | en (default: ko)
--model           Anthropic model (default: claude-opus-4-6)
--save            Persist to digests/<date>-<scope>-<window>.md

MCP server

Exposes fetch_trending(language, window) as a tool for any MCP-compatible client.

Local (stdio)

{
  "mcpServers": {
    "gh-trends": {
      "command": "uv",
      "args": ["--directory", "/path/to/gh-trends", "run", "gh-trends", "serve"]
    }
  }
}

Remote (streamable-http)

Public endpoint:

https://gh-trends.csy-p.com/mcp

{
  "mcpServers": {
    "gh-trends": {
      "type": "streamable-http",
      "url": "https://gh-trends.csy-p.com/mcp"
    }
  }
}

Stack

• Python 3.11+ with uv + pyproject.toml
• httpx — async HTTP client
• selectolax — fast HTML parsing (Lexbor)
• pydantic v2 — typed data models
• typer + rich — CLI and output formatting
• anthropic SDK — LLM-powered digest (claude-opus-4-6)
• mcp (FastMCP) — Model Context Protocol server

Development

# Install with dev dependencies
uv sync --extra dev

# Lint
uv run ruff check src tests

# Test (19 tests, no network/API calls)
uv run pytest -v

License

MIT