- Reliability 8/20
- Based only on the supplied README and repo metadata: the tool set is coherent and invocation is documented (Node 20+/npx), including a known bun incompatibility with a workaround. However, no source code, CI workflow, committed tests, or MCP manifest was available to verify the initialize/tool-list handshake or actual tool behavior. Edge cases and error handling are only inferred from the README, so I deducted from the 10 anchor because execution evidence is absent.
- Security and permissions 4/20
- No real secrets or malware indicators found; credentials are passed via environment variables and examples use placeholders only. However, the default tool set exposes salesforce_dml_records (including delete/update/upsert) and salesforce_execute_anonymous (arbitrary Apex execution), with no confirmation, dry-run, or read-only mode described in the README. This matches the red line for irreversible delete/remote-command defaults without confirmation, capping security at 4. Least-privilege scoping is also not documented.
- Maintenance 7/20
- The repository is not archived, has an MIT license, and displays an OpenSSF Scorecard badge. However, the supplied material contains no release history, commit frequency, dependency updates, issue triage evidence, or security advisory channel; 14 open issues alone do not demonstrate maintenance quality. I gave a modest score for the license and non-archived status and deducted for missing evidence.
- Documentation 15/20
- The README is layered and usable: npm/Claude Desktop/Docker setup, four auth flows with env keys and JSON examples, a full tool catalog with example prompts, known limitations (bun, API version), and development/build commands. Deductions: no parameter-level schemas, no troubleshooting/rate-limit/cost notes, and the README is untrusted evidence that cannot be cross-checked against source, so full marks are not justified.
- Setup experience 14/20
- Setup is well documented: Node 20+ requirement, global npm, manual claude_desktop_config.json examples for CLI, user/password, OAuth client-credentials and access-token flows, plus Docker build/run commands and platform caveats. Steps are concrete and copy-pasteable. Static calibration caps setup at 15 without verifiable execution evidence (real CI/tests); no runnable test output or manifest was available, so I scored 14.