| # | Repo | Lang | ★ | Tax | Size | Bloat | Agent Docs | Nav | Source | Docs | Tokens |
|---|
Champions — most agent-friendly OSS repos
The 10 lowest context-tax scores in the current snapshot. Click any card to view the detail page.
Hall of Shame — most expensive repos for AI agents
The 10 highest context-tax scores. Usually wins come from committed lock files, generated bundles, or absent agent docs.
Aggregate stats
Language distribution
Axis averages
Data sources
Every score on this site is computed at runtime from the sources listed below. The tokenizer is the public gpt-tokenizer npm package using the o200k_base encoding. No mock, seed, or synthetic data is used anywhere on this site.
| Source | URL pattern | Refresh | Last fetch | Success 7d | Avg ms 7d |
|---|
Recent fetches
| When | Kind | Target | Status | ms | Bytes |
|---|
About context-tax
Every API call to a Claude / GPT / Gemini coding agent ships the entire active context back to the model — at scale, the shape of your repo dominates the bill. context-tax is the public answer to a simple question: if two OSS libraries solve the same problem, which one is cheaper for my AI agent to understand?
For each tracked repo we:
- fetch its full recursive Git tree from GitHub,
- classify every file (source / docs / lock / generated / test / config),
- fetch up to 20 representative source files plus the README,
- tokenize them with
gpt-tokenizer(o200k_base), - compute six axes and a composite Context Tax score (0–100, lower is better).
Repo maintainers can drop the badge in their README:
Submit a repo
Submitted repos must have ≥ 50 stars and a public default branch.
Rate limit: 5 submits per hour per IP. Refresh: 1 per minute per IP per repo. All endpoints are public; no signup, no API key.