Generates source code for new FastMCP tools in Python or JavaScript, with support for dynamic resources
Single tool 'generate_mcp_tool' has critical gaps. Naming is acceptable (verb_noun pattern), but description lacks actionable context for LLM selection. Input schema is present with type declarations, but parameter descriptions are minimal (10-20 chars each) and lack constraints. No output schema documented. No error handling guidance beyond ValueError. Tool generates code as a string, not a structured, composable operation. Lacks idempotency guarantees and security considerations for code generation.
Gera código fonte para uma nova tool FastMCP em Python ou JavaScript, exibindo o resultado em preview code.
Parameter descriptions are too brief (10-20 chars) and lack format/constraint guidance. E.g., 'parameters' is described as 'Lista de parâmetros da função, por exemplo: "x: int, y: int".' but does not specify syntax rules, escaping, or validation.
No output schema documented. Tool returns a string (markdown code block), but LLMs cannot parse or validate the generated code structure. Should return structured object with {language, code, preview_url} or similar.
Tool description does not explain WHEN to use it or what happens after code is generated. Is the code executed? Saved to a file? Returned for review? Ambiguous purpose reduces LLM confidence in tool selection.
No input validation or error recovery guidance. If 'function_body' has incorrect indentation or 'parameters' uses invalid syntax, the tool raises ValueError but does not guide the LLM to fix the input. Error message should suggest corrections.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
Tool is not idempotent. Calling it twice with the same inputs produces identical outputs, but the tool does not declare this. Agents may retry on transient failures and expect deterministic behavior.