Build your own aggregator

[amacsmith]

Build your own aggregator.

A reminder that looptech-ai/understand-quickly is one implementation of CKGP, not the canonical home of code-knowledge graphs. The protocol is explicitly multi-vendor: any number of aggregators can index the same producers and serve interoperable registry.json and .well-known/repos.json endpoints.

Practical recipe to spin your own up:

1. Fork this repo. The whole site is GitHub Pages + a static registry.json + a few GitHub Actions workflows. No backend, no database.
2. Edit registry.json to point at the producers you care about — your team's repos, your stack's ecosystem, a niche language community. Validation runs locally with npm run validate.
3. Deploy. pages.yml builds on every push; sync.yml re-syncs nightly. Both are stock GitHub Actions running on the free tier.
4. Point agents at your URL. MCP clients, CLIs, and the SDK all take a registry URL. Nothing in the protocol cares whether that URL ends in looptech-ai.github.io or your own domain.

CKGP §9 ("Multi-vendor expectation") spells this out: the producer-side .well-known/code-graph.json is aggregator-independent, so any aggregator can index a producer without producer cooperation, and consumers can swap aggregators without re-publishing anything. The §9.2 anti-monopoly stance is intentional — depending on a single aggregator is the consumer's choice, not a protocol assumption.

If you spin one up, drop a link below. Curious to see what scopes (per-language, per-org, per-domain) make sense.

Spec: docs/spec/code-graph-protocol.md §9.