Introduction

github is a single binary that reads github.com over plain HTTPS and gives back records rather than pages. There is nothing to sign up for and nothing to run alongside it.

Why there is no token

The unauthenticated REST API allows sixty requests an hour. That is not enough to read one organization, let alone crawl a graph, and the moment a tool needs a token it needs a place to put it, a way to rotate it, and a story about what happens when it leaks.

The public site has none of those problems. It sits behind a CDN, so it answers faster than the API does even for the things the API covers, and it covers more: trending, contribution calendars, the dependency graph, release download counts, and the marketplace have no API route at all.

So this tool sends no Authorization header, never reads GITHUB_TOKEN or GH_TOKEN, and never touches api.github.com. A test in the package parses every source file and fails the build if any of those four strings appear outside the doctor that warns about them.

The cost is that this is read-only and public-only. For anything else, use the official gh.

Seven surfaces

GitHub serves the same facts several different ways, and no single one of them is complete. github reads all of them and treats the choice as a routing problem:

Surface	What it is
html	The rendered page, and the React payload embedded in it
route-json	The same route asked for with Accept: application/json, which is what the site's own navigation uses
xhr	Fragments the page defers and fetches separately
search	The search backend, which answers in JSON for ten types
feed	The Atom feeds for commits, releases, tags, wikis, and accounts
raw	raw.githubusercontent.com and codeload.github.com, for file bytes and archives
git	The dumb HTTP git protocol, for a complete ref list in one request

github routes prints the whole mapping, including which surface is primary for each route and which one it falls back to.

How it is built

• The gh package holds the HTTP client, the surface router, the response cache, and the typed records. It paces requests, sets an honest User-Agent, retries the transient failures any public site throws, and records for every field which URL stated it.
• The pkg/page package does the extraction: the React payload, the preloaded Relay queries, schema.org blocks, ld+json, og and twitter meta, microdata, and the deferred fragments a page names for itself.
• The cli package wraps operations in subcommands.
• One cmd/github entry point ties them together.

Operations are declared once with kit and appear as a CLI command, an HTTP route, an MCP tool, and a github:// URI type without any per-surface code.

Next: install it, then take the quick start.