- Reliability 5/20
- Limited evidence: README claims MCP is off by default, mounts POST /mcp when MCP_ENABLED=true, and exposes 51 curated tools behind API-key auth; but no server manifest, source, tests, or CI execution evidence is supplied, so the declared tool list cannot be matched to real behavior and error paths/dependencies are not diagnosable. Deduction: static calibration requires verifiable execution evidence and missing evidence lowers the score.
- Security and permissions 10/20
- No red lines triggered (no malware, credential theft, real secrets, or default destructive actions). README documents a solid control surface: MCP disabled by default, dedicated least-privilege keys, OPERATOR/session scoping, MCP_READONLY, rate limiting, and an explicit warning not to expose /mcp publicly; examples use placeholders only. Deduction: these controls are only documented, not verifiable from source, and explicit confirmation for dangerous operations is not present in the material.
- Maintenance 6/20
- Evidence shows the repository is not archived, has an MIT license, and few open issues; the README also references CI and a docs tree. These are surface activity signals. Deduction: no verifiable commit history, releases, dependency-update cadence, or security-response channel, so sustained maintenance and governance cannot be confirmed.
- Documentation 14/20
- The supplied README is well layered: install steps, ports, curl API examples, MCP .mcp.json client config, auth/read-only/rate-limit guidance, WhatsApp ban-risk warnings, security architecture, and a docs index. Deduction: the 51 MCP tools and their parameter schemas are not listed, the actual docs files are not supplied, and there is no evidence of more granular cost or troubleshooting guidance.
- Setup experience 12/20
- The install path is clear: one-command Docker Compose, local npm ci flow, production profiles, port table, MCP client configuration example, and multi-arch images. Deduction: MCP is off by default and API-key creation is not shown; the WhatsApp session/QR flow is not independently verifiable; under static calibration, missing test/CI evidence caps this dimension at 15.