An MCP server implementing the think tool for Claude and other LLMs
Single tool 'think' with minimal quality markers. Tool is explicitly registered with schema visible in src/server.ts (line 26-38). Description present but generic and not LLM-optimized. Input schema uses Zod and maps to JSON Schema, but lacks richness expected of production tools. No output schema documented. No error handling guidance. No parameter constraints beyond type. Tool does minimal work, accepts a string thought, returns truncated echo. Lacks idempotency markers, error recovery, and composition signals.
Use this tool to think about something. It will not obtain new information or change anything. Use it when complex reasoning is needed. Args: thought: A thought to think about. This can be structured reasoning, step-by-step analysis, policy verification, or any other mental process that helps with problem-solving.
Tool name 'think' lacks action verb convention. Does not start with get_, create_, update_, delete_, search_, list_, send_, etc. Single-word names make intent ambiguous for LLM tool selection.
Description is 237 characters but lacks clarity on WHEN to use this tool vs alternatives, WHAT happens when called, and WHAT gets returned. Generic text like 'Use this tool to think' does not guide LLM selection. Missing dependency hints (e.g., 'Call this when reasoning through ambiguous cases').
No output schema documented. Response hardcoded as text truncation (src/server.ts line 35-38) but LLM has no schema to plan downstream calls or extract structured data.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 44 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 37 | - | v1 |
Input parameter 'thought' description is identical to tool description, duplicated boilerplate text. Does not clarify acceptable formats, length limits, or structure types (e.g., 'step-by-step analysis' vs 'policy verification').
No constraints or validation. Parameter 'thought' is unbounded string, LLM could pass megabytes of text. No min/max length, no format regex, no enum. Tool implementation truncates at 50 chars silently (line 36-37), but LLM receives no error if input is invalid.
No error handling or recovery guidance. Tool always returns success with truncated text. What if thought parameter is empty string? What if LLM passes null or very large payload? No error classification (retryable/user-fixable/fatal). Per pattern, error responses must tell LLM what to do next.
Tool annotation hints missing. No readOnlyHint, destructiveHint, or idempotentHint declared. Per current spec (2026-07-28), tool annotations enable LLMs to reason about side effects. 'think' is read-only and idempotent but these are not marked in schema.