GitHub AI Radar MCP

A small stdio MCP server for discovering AI-related GitHub repositories.

The first version focuses on GitHub Search API signals, with optional GH Archive growth enrichment:

• Search AI-related repositories by topic and keyword.
• Find repositories created or pushed on a date.
• Return a Top 20 AI repo radar list with an approximate attention score.
• Optionally count same-day stars/forks/issues/PRs from GH Archive for candidate repositories.
• Discover trending AI repositories directly from GH Archive global events.

GH Archive dates and hours are UTC.

Requirements

• Node.js 20+
• Optional: GITHUB_TOKEN or GH_TOKEN

GitHub unauthenticated search is heavily rate-limited. A token is strongly recommended.

Run

npm start

Generate a Markdown report directly from the command line:

npm run report -- --date 2026-05-05 --hours 0,1,2,3 --candidate-limit 150 --limit 20

Save to a specific file:

npm run report -- --date 2026-05-05 --hours 0,1,2,3 --output reports/github-ai-radar-2026-05-05.md

Create a local config from the example:

cp config.example.json config.json
npm run report

For MCP clients, configure stdio like this:

{
  "mcpServers": {
    "github-ai-radar": {
      "command": "node",
      "args": ["/Users/pengfei.chen/Documents/New project/src/server.js"],
      "env": {
        "GITHUB_TOKEN": "ghp_xxx"
      }
    }
  }
}

If installed from npm, configure the package binary instead:

{
  "mcpServers": {
    "github-ai-radar": {
      "command": "npx",
      "args": ["-y", "github-ai-radar-mcp"],
      "env": {
        "GITHUB_TOKEN": "ghp_xxx"
      }
    }
  }
}

Tools

github_ai_search_repos

Search AI repositories using GitHub Search API.

Example arguments:

{
  "query": "agent framework",
  "topics": ["ai-agent", "llm", "mcp"],
  "pushed_from": "2026-05-05",
  "sort": "stars",
  "limit": 20
}

github_ai_repos_today

Find AI repositories created or pushed on a date.

Example:

{
  "date": "2026-05-05",
  "mode": "updated",
  "limit": 20
}

github_ai_top20_today

Return a Top 20 radar list. By default it uses GitHub Search API only. Pass use_gharchive: true to enrich candidate repositories with exact same-day event counts from GH Archive.

Example:

{
  "date": "2026-05-05",
  "topics": ["ai-agent", "llm", "mcp"],
  "include_readme": false,
  "use_gharchive": true,
  "archive_hours": [0, 1, 2, 3]
}

Scanning all 24 GH Archive hours can take time and download a lot of data. Use archive_hours while testing.

github_ai_growth_today

Scan GH Archive for daily GitHub activity on AI candidate repositories.

Example:

{
  "date": "2026-05-05",
  "topics": ["ai-agent", "llm", "mcp"],
  "candidate_limit": 50,
  "archive_hours": [0, 1, 2]
}

github_ai_trending_today

Discover trending AI repositories from GH Archive global events.

This is the most useful tool for the original goal: "today's most watched AI repositories on GitHub."

Flow:

GH Archive global Watch/Fork/Issue/PR events
  -> rank active repositories
  -> fetch GitHub repo metadata
  -> classify AI-related repos
  -> return Top N

Example:

{
  "date": "2026-05-05",
  "archive_hours": [0, 1, 2, 3],
  "candidate_limit": 150,
  "limit": 20,
  "include_readme": false
}

Use include_readme: true when you want richer AI classification, but it costs extra GitHub API calls.

github_ai_report_today

Generate a Markdown report using the same GH Archive trend discovery path.

Example:

{
  "date": "2026-05-05",
  "archive_hours": [0, 1, 2, 3],
  "candidate_limit": 150,
  "limit": 20,
  "save_to_file": true
}

The returned payload includes both structured repos and a markdown field. When save_to_file is true, it also returns saved_path. The default path is:

reports/github-ai-radar-YYYY-MM-DD.md

Prompts

github_ai_daily_brief

An MCP prompt template that asks the client to generate a concise Chinese daily brief using github_ai_report_today.

Returned repos include:

{
  "today_stars": 12,
  "today_forks": 2,
  "today_issues": 1,
  "today_prs": 3,
  "today_events": 18
}

Scoring

Without GH Archive, the score is approximate:

score = total stars + forks + open issues + AI topic matches + pushed-today boost

With GH Archive enabled, same-day events become a major part of the score:

daily_score = today_stars * 5 + today_forks * 3 + today_issues + today_prs

The returned payload includes a score_note so downstream AI clients know whether exact daily event counts were used.

Cache

GH Archive hourly files, GitHub repository metadata, README excerpts, and GitHub Search responses are cached under:

.cache/github-ai-radar/

Override with:

GITHUB_AI_RADAR_CACHE_DIR=/path/to/cache npm start

Next Steps

1. Add SQLite cache for more advanced expiration and indexing.
2. Add optional dashboard if needed.
3. Add scheduled automation wrapper if needed.