Data handling

The project website is static. Its course examples are fictional, and it does not connect to a student account or accept credentials. GitHub Pages still has its own hosting and access-log behavior. Public documentation omits private account, institution and local-machine details from validation reports.

The connector has no telemetry, analytics, advertising SDK, model SDK or hosted Folava backend. It makes requested Canvas reads, attachment requests through API pointers and S3, and explicit credential lifecycle requests. Normal DNS, TLS and Canvas access logs still exist. Canvas may also refresh derived progress caches during GET requests; the connector does not promise that remote account state is unchanged. The validation record explains the unresolved sampled module-requirement signal.

Your MCP host sends selected tool inputs and Canvas results to the model provider you use. Its storage, training and retention terms apply separately. This package cannot promise that another provider will not retain or train on that content. Use only a host and provider appropriate for your institution and the data involved.

Credentials are stored in the OS keyring, or supplied by the operator's environment or protected token file. Metadata records the Canvas origin, user ID, OAuth scope set and non-secret references. Polling stores hashes, query scopes, resource IDs and event times/kinds. These are private metadata; hashes are not anonymization. Exports store file bytes at a path you explicitly choose. MCP/API results are not persisted by this package, but the host may keep transcripts and raw signed URLs.

Use auth logout to remove a stored profile and keyring credential, with --revoke when you intend remote revocation. Delete your event SQLite databases and exports when no longer needed. Remove injected secrets from their original source. Secure backups, transcripts and credential-store retention are your operator's responsibility.

This document describes this package's behavior; it is not a policy for a hosted commercial service. Such a service would need its own data terms and verified institutional/API approval.