Loading ToolBench
A Go-based MCP server that manages tool catalogs, capability tools, and agent runtime with support for task management and user input handling
Single tool 'ask_input' has a well-structured JSON Schema with proper type definitions, required fields, and pattern constraints. Description is clear and actionable (145 chars), explaining the tool's purpose and parameter semantics. However, the tool is narrowly scoped (user input elicitation only) and lacks error handling guidance. No output schema is documented. The server appears to be a specialized agent runtime component rather than a general-purpose tool provider, limiting its applicability.
Pause the current task only when the typed outcome contract or a structured tool failure says user input is required. The nonblank question field is authoritative. Use choices=[] for free-form input, or provide choices to let the user pick one of them or type a different answer.
Missing output schema documentation. Tool description does not specify what the response structure looks like (e.g., does it return {answer: string}? {selected_choice: string}?).
No error handling guidance. Description does not explain what happens if user cancels, times out, or provides invalid input. LLM has no recovery path.
Limited tool scope. Server provides only one tool (ask_input) for user interaction. No complementary tools for other agent patterns (tool composition, chaining).
Parameter 'choices' lacks minItems/maxItems constraints. Unbounded array could allow LLM to pass thousands of choices, wasting tokens.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2025-06-18+ | v2 |