Uploads and MCP evidence
Safe archive ingestion and the OAuth-protected MCP evidence gateway for structured evidence and raw artifacts.
File uploads
Uploads enter the same ingestion pipeline as GitHub snapshots but have stricter provenance. The service enforces compressed and unpacked size limits, file-count and decompression-ratio limits, path and symlink traversal protection, MIME inspection, and malware scanning.
No uploaded hook, script, binary, package lifecycle command, build, test, or customer source is executed.
MCP gateway
The Streamable HTTP MCP service exposes a narrow evidence-ingestion surface:
submit_evidence_batchcreate_artifact_uploadcomplete_artifact_uploadget_ingestion_status
It uses OAuth Authorization Code with PKCE S256, audience-bound access tokens, resource indicators, single-use authorization codes, rotating refresh tokens, and explicit browser consent with workspace selection.
Supported scopes are evidence:write, uploads:write, and ingestion:read.
Evidence acceptance
Structured batches are schema-validated, tenant-bound, provenance-stamped, and normalized before reconciliation. Raw artifacts use scoped signed uploads and pass through malware and archive-safety checks. Neither is passed directly to an agent prompt.
MCP is an evidence gateway
Hyperoru does not execute arbitrary outbound MCP tools as part of an audit. The integration accepts evidence; it does not grant agents a general remote-action surface.