- Reliability 8/20
- The manifest and README contradict each other on total tool count: the object description says '875 tools', server.json's own description says '890 tools in 13 profiles, narrow to 27', and the README feature table says 'monetization:subscription-pricing is 24 tools instead of 204' — three different sets of numbers for what should be the same generated surface, suggesting the tool manifest and docs have drifted out of sync. No CI workflow or test files are included in the provided material to verify the MCP init/tool-list handshake or the behavior of the 875+ generated tools, so per the static calibration rule reliability is capped at 12; the actual score is further reduced for the unresolved numeric contradictions and total absence of verifiable execution evidence.
- Security and permissions 14/20
- The README states that write operations require confirmation by default (must explicitly set ASC_CONFIRM_WRITES=0 to disable), offers a --read-only mode, marks destructive actions, pins requests to a host, claims no telemetry, and stores the private key in the macOS Keychain rather than a plaintext config, with the private-key env var explicitly flagged isSecret — this reflects a reasonable least-privilege, secure-by-default design, and no real secrets/tokens appear in any example, so no red line is triggered. However, all of this is self-reported in the README/manifest with no source code or the actual SECURITY.md content provided to verify that the confirmation logic, host-pinning, or Keychain handling are genuinely implemented, so full marks are not warranted.
- Maintenance 12/20
- The project uses the MIT license, is at version 2.3.0, and the README references CHANGELOG.md, GOVERNANCE.md, CONTRIBUTING.md and SECURITY.md, and documents how the 2.0.0 breaking change was communicated — indicating some versioning discipline. The repo is not archived and has 0 open issues. But there are signs of a single maintainer (a personal donation link), no actual commit history, release cadence, or issue-response evidence is provided, and no CI/automation material is shown; star count is explicitly excluded from scoring. This nets out to moderate maintenance activity with limited supporting evidence.
- Documentation 14/20
- The documentation set is fairly extensive: a bilingual (English/Turkish) README linking to GUIDE, SECURITY, SUPPORT, CHANGELOG, CONTRIBUTING, CODE_OF_CONDUCT, GOVERNANCE and AGENTS docs, covering install, auth, configuration examples, safety modes, and multi-client registration. However there are clear internal inconsistencies (875 vs 890 total tools; 27/207 vs 24/204 sub-profile narrowing examples) that undercut confidence in the documentation's accuracy, and the content of the linked detailed docs itself is not provided for verification.
- Setup experience 13/20
- The manifest is well-structured: npm package, stdio transport, an optional positional profile argument, and explicit environment variables with required/secret flags. The README gives a single `npx -y @erayendes/asc-mcp setup` command and describes it registering profiles across multiple clients, which reads as a low-step, clear installation path in principle. But with no executable evidence (no CI, no recorded install test) to confirm the setup wizard's real-world behavior, setup is capped at 15 per the static calibration rule, and the score is further tempered by the lack of any verifiable execution record.