Loading ToolBench
MCP server providing a 'think' tool for structured reasoning, inspired by Anthropic's Claude 'think' tool. Enables agentic LLMs to pause, log thoughts, and improve multi-step tool use.
Server provides 4 tools with minimal quality. All tools have descriptions (10-50 chars, below 194-char baseline), but lack depth and actionable guidance. Input schemas are present but sparse: each tool accepts only a single string parameter with minimal constraint. No output schemas documented. Tools are read-only stubs that echo input rather than perform meaningful work. The 'think', 'criticize', and 'plan' tools are nearly identical in purpose and description, violating single-responsibility principle. 'search' tool lacks error handling, timeout guidance, and output structure documentation. No parameter validation, no error recovery guidance, no security considerations.
Use the tool to critic your steps. It will not obtain new information or change the database, but just append the thought to the log. Use it when complex reasoning or some cache memory is needed.
Use the tool to plan your steps. It will not obtain new information or change the database, but just append the thought to the log. Use it when complex reasoning or some cache memory is needed.
Search the web for a given query.
Use the tool to think about something. It will not obtain new information or change the database, but just append the thought to the log. Use it when complex reasoning or some cache memory is needed.
Three tools (think, criticize, plan) are functionally identical, all accept a string, echo it back, and claim to 'append to log'. This violates single-responsibility principle and forces LLM to guess which to use.
No output schemas documented. LLMs cannot infer what 'search' returns (JSON structure, fields, pagination). Baseline: 100% of A+ tools document return types.
Descriptions are 30-50 chars (baseline 194 chars, p10=34, p90=392). Too terse to guide LLM selection. E.g., 'Use the tool to think about something' lacks WHEN to use it, WHAT it returns, or prerequisites.
search tool has no error handling, timeout guidance, or recovery instructions. If Tavily API fails or times out, LLM receives raw exception with no actionable next step.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | 2026-07-28+ | v2 |
No parameter constraints (enums, min/max, regex patterns). 'query' and 'thought' accept any string, inviting hallucinated or malformed input. Baseline: constrained inputs prevent invalid LLM submissions.