Loading ToolBench
AI Extension that acts as an MCP client for thinking-partner-mcp server.
Three tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. All tools have descriptions (10-60 chars, below the 50-200 char baseline for LLM optimization). Input schemas present but parameters lack depth. No error handling guidance, no output schema documentation, and no tool annotations. The server acts as a thin MCP client wrapper around an external thinking-partner-mcp server, which limits its ability to provide rich, self-contained tool definitions.
Retrieve current focus and context information from the thinking-partner-mcp server
Retrieve the focus history from the thinking-partner-mcp server
Set a new focus topic with optional context for the thinking-partner-mcp server
Descriptions are too short (10-60 chars vs. 50-200 char baseline). 'Retrieve current focus and context information' lacks WHEN to use, WHAT it returns, and prerequisites. LLMs cannot reliably select tools with such sparse guidance.
No output schema documented. Callers must infer that responses are JSON with 'success', 'context', 'focus', or 'currentFocus' fields. LLMs cannot plan downstream calls or extract data reliably without explicit return type documentation.
Parameter descriptions are minimal. 'The tool context (e.g., desktop or code)' does not explain WHY this parameter exists, WHEN to use each value, or what happens if omitted. 'The scope of context to retrieve' lacks detail on the difference between 'current' and 'all'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
No error handling guidance. Code catches errors and returns fallback values (e.g., { currentFocus: null, recentContext: [] }) but does not tell the LLM what to do next. If a call fails, the agent has no recovery path.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). get_context and get_focus_history are read-only; set_focus is destructive. Agents cannot infer safety properties without explicit annotations.