Loading ToolBench
This server has 2 tools with basic schemas and descriptions, but significant quality gaps prevent production use. The 'generate' tool has a minimal description (28 chars) and sparse parameter documentation. The 'generate_plan' tool description is incomplete (truncated in source). Both tools lack output schema documentation. Parameter descriptions are minimal. Error handling exists but is not well-integrated into tool definitions. No guidance for LLM recovery paths. The server embeds credentials (GEMINI_API_KEY) as environment variables but does not document this in tool schemas.
Generates text using Gemini.
Generates a project plan structure from a goal.
generate_plan tool description is truncated/incomplete in source code. The description field is cut off and does not provide sufficient context for LLM tool selection.
No output schemas documented. Neither tool specifies what fields or structure the LLM should expect in the response. This prevents agents from planning downstream actions or extracting relevant data.
Parameter descriptions are minimal. The 'prompt' parameter in 'generate' is described as 'Prompt to generate text from' (24 chars), lacks context on format, length limits, or expected structure. The 'goal' parameter in 'generate_plan' has no visible description.
Tool names 'generate' and 'generate_plan' are generic and do not follow verb_noun naming convention. 'generate' is ambiguous, does it generate text, code, images? Should be 'generate_text'. 'generate_plan' lacks specificity, generate_project_plan would be clearer.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 10 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 31 | - | v1 |
No error recovery guidance. Error responses raise McpError with INTERNAL_ERROR codes but do not tell the LLM what to do next (e.g., 'Try with a shorter prompt' or 'Check that the goal is specific and measurable'). Agents receive no actionable remediation path.
generate tool exposes external API dependency (Gemini) without documenting rate limits, quota constraints, or failure modes. LLMs will not know when to back off or request user intervention.
generate_plan tool internally spawns a subprocess to call gemini-server. This subprocess pattern is fragile, undocumented, and not visible to the LLM. If the gemini-server subprocess fails, errors are opaque to the agent.