Loading ToolBench
The boilerplate provides a well-structured HTTP MCP server with one tool ('view') that has clear input/output schemas and proper annotations. The tool definition demonstrates good naming conventions, comprehensive schema documentation, and proper error handling with recovery guidance. However, the server is a boilerplate/framework rather than a feature-complete application, limiting the depth of evaluation. The single tool exhibits professional-grade definition quality with actionable error messages and structured output. Input schemas use Zod validation with proper type definitions and descriptions.
View Skill
Tool name 'view' is too generic and lacks action verb clarity. Does not follow verb_noun pattern. Should be 'view_skill' or 'get_skill_content' to clearly signal the action.
Parameter 'skill_name' accepts '.' as a special sentinel value to list available skills. This overloads the parameter semantics and makes the tool's behavior non-obvious. Better approach: provide separate tool 'list_skills' or use a dedicated boolean parameter 'list_only=true'.
The tool description states it views 'Skill' but does not document what a 'Skill' is in the context of this system. New users cannot determine whether this tool is appropriate without external context. Add context: 'View contents of a Skill (a structured knowledge package) stored in [system].'
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-21 | B | 72 | 2026-07-28+ | v2 |
| 2026-03-09 | B | 70 | - | v1 |
Error handling returns plain text messages in a 'content' field rather than structured error objects. While the messages are helpful, a properly structured error response (e.g., with error_code, recovery_action fields) would be more machine-parseable for sophisticated agents.
Output schema returns all results as unstructured text in a single 'content' field. While appropriate for a generic skill viewer, listing many skills (via '.') returns a potentially very long plain-text response. No pagination or result limits documented in the tool description.