GitHub Repository Search and New Repo Monitor

Search every public repository on GitHub by topic, language, stars or creation date. Or run it on a schedule and get only the ones that appeared since last time.

No key and no login. A GitHub token is accepted and is not needed.

What makes this one different

It tells you what you did not get. GitHub returns at most 1,000 results for any one search, however many there are. A search for model context protocol server matches over 100,000 and hands back a thousand of them without saying so. Every row here carries three fields: how many repositories match on GitHub, how many this run returned, and whether those two agree. If they do not, you know, and the file is still useful.

It goes past the ceiling when you ask it to. Set a result limit above 1,000 and it splits the creation date range into windows and works through them, because a narrower window fits under the cap. That is the only way to enumerate a large query on GitHub.

One thing to know about that: inside a single window GitHub does the ranking, but across windows it cannot. So asking for the top 200 by stars gives you the real top 200. Asking for 1,200 gives you 1,200 real matches with the top of each date window among them, not a global top 1,200. GitHub has no way to give anyone the second thousand of a ranking.

It never guesses at a rate limit. GitHub allows 10 searches a minute without a token, and going over returns a 403 that also costs you the next window. This waits for the reset rather than retrying into it, and says on the run how long it waited.

What you get

One row per repository: owner, name, description, homepage, main language, topics, stars, forks, licence, default branch, size, and three dates. Created, updated and last push.

Read pushedAt for whether a project is alive. updatedAt moves when someone stars the repository, so it is not a sign of activity.

Two fields that are usually reported wrong

Open issues include pull requests. GitHub counts a pull request as an issue in its search results and offers no separate issue number. On apify/crawlee the field reads 152, which is 109 open issues and 43 open pull requests. The column here is called openIssuesAndPullRequests for that reason.

There is no watchers column. In a search result GitHub's watchers_count is not watchers, it is the star count again, exactly. The real number lives on a different endpoint that allows 60 requests an hour, so an Actor that fetched it would truncate at 60 repositories. Leaving the column out is honest. Publishing stars twice under two names is not.

Monitoring

Turn on Only new since the last run and schedule it. The first run records a baseline and returns everything. After that you get only repositories that were not there before, marked isNew.

A query whose request fails is carried forward untouched, so a rate limit or a GitHub outage cannot look like repositories vanishing. Nothing is ever reported as removed, because a repository leaving a result set means the ranking moved or it was renamed.

The token

Without one you get 10 searches a minute, which is 1,000 repositories a minute. A personal access token with no scopes at all raises that to 30 a minute. Nothing else changes, and the results are identical.

Limits worth knowing before you buy

• 1,000 results per search window, set by GitHub. Splitting the date range is the way around it and the Actor does that for you.
• Forks are excluded unless you ask for them, which is what GitHub's own search does.
• Private repositories are never included, with or without a token.
• Contributor counts, README text and release data are not here. They need one request per repository against a 60 an hour limit, so they belong in a different Actor with its own honest ceiling rather than silently capping this one at 60 rows.