- Reliability 12/20
- The project uses MCP SDK v2 with clear protocol negotiation for version 2026-07-28 while maintaining legacy client compatibility. The code structure is well-organized with complete implementations of tools, prompts, and resources. However, there's insufficient evidence of thorough error handling and edge case testing, preventing a full score.
- Security and permissions 16/20
- The project employs a security-first design, defaulting to not return sensitive data unless explicitly requested with reveal: true. It supports op_run functionality for injecting environment variables into child processes without exposing plaintext in model outputs. It also integrates with macOS Keychain and provides allow-list configuration. However, the MCP protocol itself doesn't encrypt secrets in transit, presenting some risk.
- Maintenance 14/20
- The project has active CI/CD processes with frequent version updates, with the latest being 4.0.2. It uses the Apache 2.0 license and includes clear contribution guidelines. However, dependency update status and issue response times cannot be determined from static code alone.
- Documentation 18/20
- Documentation is comprehensive, including detailed installation guides, configuration options, tool descriptions, and usage examples. It particularly emphasizes security best practices, including how to handle sensitive information and follow the principle of least privilege. Configuration examples are provided for various clients like Claude Desktop, VS Code, and OpenAI Codex.
- Setup experience 15/20
- Installation is straightforward using npm package management, supporting configuration via environment variables or CLI parameters. It provides configuration guides for different platforms, especially macOS Keychain integration. Configuration examples are clear and easy to understand and implement.