Transparency
How you can check what this service runs and what happens to your data.
1. Which code is running
The MCP server, the rate-limit gateway, the API-key service, the Korean tools and the document generator are deployed only by GitHub Actions from workers/. Each deploy bundles that folder, records a build provenance attestation, and stamps the live service with the commit and bundle hash.
Verify it yourself:
git clone https://github.com/etacbae6768-a11y/tools-that-dont-upload && cd tools-that-dont-upload git checkout <commit above> tar --sort=name --mtime='2020-01-01' --owner=0 --group=0 --numeric-owner -czf workers-bundle.tgz workers sha256sum workers-bundle.tgz # should match bundle_sha256 above gh attestation verify workers-bundle.tgz --repo etacbae6768-a11y/tools-that-dont-upload
If the live service reports deployed_via: manual (not attested), the running code was changed outside CI and is not covered by an attestation. Other parts of the site (static pages, older utility workers) are not yet under CI.
2. Where your data goes
- Every API response has an
X-Egressheader and every MCP result has_meta["com.toolsthatdontupload/egress"]listing external hosts contacted. For most tools it isnone. - Each operation in /openapi.json lists its possible egress in
x-egress. - Request bodies are not stored or logged by us. Details: Privacy Policy.
3. What it costs us to run (measured)
Cloudflare Workers analytics, 2026-10-05 to 2026-10-12 (36,934 requests across all workers). Prices: Workers Paid, $0.30 per million requests and $0.02 per million CPU-ms.
| Call type | CPU time (p50 / p99) | Infra cost per 1,000 calls | Credits charged (keyed) |
|---|---|---|---|
| Compute tool (REST) | 0.62 ms / 3.13 ms | $0.00031 | not metered yet |
| Compute tool via MCP | +0.66 ms | $0.00063 | not metered yet |
| Network tool (gateway + fetch) | +1.39 ms | $0.00064 + outbound time | 20 credits |
| AI tool (Workers AI) | 1.27 ms (worker only) | worker cost + Workers AI neurons (billed separately by Cloudflare) | 50–100 credits |
Credits are priced to cover outbound requests, AI inference and abuse handling, not just CPU time.
4. Service status
Live checks every 15 minutes: /status/.