gh-momentum 0.1.0

Surface GitHub repos gaining traction fast, before they hit the front page.

gh-momentum

Surface GitHub repos gaining traction fast — before they hit the front page.

gh-momentum queries the GitHub Search API for newly-created repositories and ranks them by momentum (stars per day), not by total stars. A repo with 800 stars that's 6 days old is a stronger signal than one with 40,000 stars that's 8 years old. This tool finds the former.

Zero dependencies. Pure Python standard library. One file of logic, one CLI.

$ gh-momentum --days 7 --min-velocity 50

   9.1/10  acme/turbo-agent  (1240*, 177.1/day, Python)
          A minimal autonomous agent runtime that fits in 500 lines.
          https://github.com/acme/turbo-agent

   7.4/10  data-co/ducklake  (612*, 87.4/day, Rust)
          Embedded analytics database with a Postgres wire protocol.
          https://github.com/data-co/ducklake

Why

Star count is a lagging indicator. By the time a repo has 20k stars, the opportunity to be early — to contribute, to build on top of it, to write the first tutorial — is mostly gone. Star velocity is a leading indicator. gh-momentum ranks by velocity so you see things while they're still small.

Install

pip install gh-momentum

Or from source:

git clone https://github.com/Dheightsan/gh-momentum
cd gh-momentum
pip install -e .

Usage

# Repos created in the last 7 days, gaining at least 50 stars/day
gh-momentum --days 7 --min-velocity 50

# Also pull in repos tagged with specific GitHub topics
gh-momentum --topic llm --topic rust

# Boost repos that match keywords you care about
gh-momentum --match "python,fastapi,cli"

# Machine-readable output
gh-momentum --json | jq '.[] | .name'

Rate limits

The GitHub Search API allows 60 requests/hour unauthenticated. Set a token to raise that to 5,000/hour:

export GITHUB_TOKEN=ghp_your_token_here
gh-momentum
# or: gh-momentum --token ghp_your_token_here

A token with no scopes at all is enough — gh-momentum only reads public data.

Use as a library

from gh_momentum import find_momentum

for repo in find_momentum(days_back=7, min_velocity=50, match=["llm"]):
    print(repo.score, repo.name, repo.star_velocity_per_day)

find_momentum() returns a list of MomentumRepo dataclasses, sorted by score.

MCP server

gh-momentum ships an optional Model Context Protocol server, so AI agents can query GitHub momentum as a tool.

pip install "gh-momentum[mcp]"
gh-momentum-mcp          # runs over stdio, for MCP clients

It exposes one tool, find_trending_repos, with the same parameters as the CLI. Point any MCP-compatible client at the gh-momentum-mcp command. The core CLI and library stay dependency-free — only the MCP server pulls in mcp.

How the score works

The 0-10 score combines three signals:

Signal	Weight
Star velocity	Primary. stars/day, saturates near 8.0
Absolute stars	Small confidence bonus (up to +1.0)
Keyword match	Optional boost when --match keywords hit (up to +1.5)

The scoring lives in one function — _score() in gh_momentum/detector.py — so it's easy to read, fork, and tune to your own taste.

Development

pip install -e ".[dev]"
pytest -q

Tests run fully offline. The one network test is opt-in:

REPO_RADAR_LIVE_TEST=1 pytest -q

Contributing

Issues and PRs welcome. This is maintained on a best-effort basis — if you need a feature, a PR is the fastest path. Keep the zero-dependency rule intact.

License

MIT — see LICENSE.