Loading ToolBench
MCP (Model Context Protocol) server bridge for the Praxis tool engine. Exposes canonical Praxis tools (filesystem, editing, shell, memory) over the Model Context Protocol.
Two tools with minimal definitions. 'echo' has a basic schema with one parameter; 'fail' has an empty schema. No parameter descriptions provided for 'echo'. No output schemas documented. No error handling guidance. No composition or chaining support. This server exposes toy tools that do not meet production quality baselines.
Echoes the input message
Always fails
Tool descriptions are critically short (under 20 chars). 'echo' = 'Echoes the input message' (24 chars, borderline); 'fail' = 'Always fails' (12 chars). Neither tool explains WHEN to use it, WHAT it returns, or any prerequisites.
'fail' tool has NO input schema (empty object). This tool is non-functional for agent composition, it cannot accept parameters and provides no guidance on failure modes.
'echo' parameter 'message' lacks a description. LLMs cannot infer that 'message' is the text to echo vs. a message ID or format specifier.
No output schemas documented for either tool. LLMs cannot plan downstream calls or extract structured data. 'echo' presumably returns the echoed text, but this is not declared. 'fail' provides no guidance on error structure.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 31 | 2025-06-18+ | v2 |
No error handling guidance. 'fail' is designed to fail, but the tool provides no recovery hints, error classification, or actionable next steps for the agent. A bare failure message wastes agent reasoning cycles.