- Reliability 10/20
- The README describes the main MCP initialization, aggregation, forwarding, SSE, and Streamable HTTP paths, making the happy path plausible. However, no cached server manifest, execution evidence, committed tests, or CI coverage is provided; some middleware capabilities are marked as forthcoming, so error handling, edge behavior, and actual tool consistency cannot be verified. Score reduced accordingly.
- Security and permissions 12/20
- The documentation shows API-key and Bearer authentication, OAuth/OIDC with PKCE, multi-tenancy, private/public scopes, and rate limiting, and recommends environment-variable references for secrets. However, the configurable gateway launches STDIO commands and connects to external MCP services; privilege isolation, dangerous-operation confirmation, data boundaries, and secret-leak protections are not sufficiently evidenced. Rate limits are also process-local in clusters, so the score is limited.
- Maintenance 10/20
- The project has an MIT license, a repository and contribution path, and the README mentions continuing to merge PRs. However, the author explicitly notes recent maintenance delays, and the supplied material does not verify release cadence, commit activity, dependency updates, vulnerability response, or a clear maintenance commitment. Missing evidence is deducted.
- Documentation 15/20
- The README covers Docker Compose, local development, client connections, authentication, OIDC, rate limiting, logging, proxy configuration, cold starts, and stated limitations, with several configuration examples. It lacks complete tool parameters, cost information, data-flow boundaries, dangerous-side-effect confirmation details, and reviewable test evidence; some capabilities are explicitly marked coming soon. Score reduced for these gaps.
- Setup experience 13/20
- Docker Compose, environment preparation, dev containers, local development, and Cursor, Claude Desktop, and proxy configurations are documented, making the main connection path reasonably clear. However, Docker, Postgres, external network access, possible custom dependencies, and reverse-proxy configuration may be required; STDIO-only clients need an additional proxy, so setup is not low-friction. No CI or test evidence supports a higher score.