Loading ToolBench
A server for the prompt engineer MCP that rewrites user's coding prompts before passing to AI IDE (e.g. Cursor AI) to get the best results from AI IDE.
Single tool with a clear verb-based name ('rewrite_coding_prompt') and reasonable description (105 chars). Input schema is complete with proper JSON Schema structure and parameter descriptions. However, the tool lacks critical production patterns: no output schema documentation, no error recovery guidance, no pagination support (acceptable for this use case), no parameter validation constraints, and no addressing of LLM-friendly identifiers. The implementation uses Claude API internally but does not document this dependency or the cost/latency implications. Error handling returns raw exception messages rather than actionable recovery paths. The server is feature-complete for its narrow scope but exhibits gaps in enterprise-grade tool design.
Rewrites user's coding prompts before passing to AI IDE (e.g. Cursor AI) to get the best results from AI IDE.
No output schema documented. The tool returns a string (rewritten prompt) but LLMs cannot verify the structure or plan downstream usage. Production tools should declare: 'Returns: {type: "object", properties: {rewritten_prompt: {type: "string", description: "The optimized prompt for Cursor AI"}}}' or similar.
Error handling does not guide recovery. When the tool fails (API key missing, Claude API timeout, invalid response), it returns raw error strings like 'ANTHROPIC_API_KEY environment variable is not set' or 'Invalid response from Claude API'. Per pattern:recovery-guide, errors should tell the LLM what to do: 'ANTHROPIC_API_KEY not set. Please configure this environment variable in the MCP server settings before calling this tool again.'
Parameter 'language' lacks validation constraints. Description says 'The programming language of the code' but does not specify: valid languages (TypeScript, Python, Java, etc.), whether it is case-sensitive, or what happens if an unsupported language is passed. LLMs will hallucinate values like 'JAVASCRIPT' or 'js' without formal constraints (enum or regex pattern).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 67 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 39 | - | v1 |
Tool description does not document prerequisites or dependencies. It does not mention that the tool requires a valid ANTHROPIC_API_KEY environment variable, depends on Claude 3.5 Sonnet, or may incur API costs. Per pattern:tool-description, descriptions should state WHAT, WHEN, and any prerequisites.
Tool does not declare which operations are safe to retry and which are not. The rewrite operation is idempotent (calling it twice with the same inputs produces the same output), but this is not documented. Per pattern:idempotent-operation, agents need this info to know whether a retry after timeout is safe.