Loading ToolBench
CI-grade evidence and safety tools for AI agents, MCP servers, and open-source contribution workflows.
This is a test fixture server, not a production MCP server. All 4 tools are named 'echo' with identical or near-identical definitions. Tool naming violates the verb_noun pattern (echo is a verb but lacks a noun object). Descriptions are minimal (10-60 chars) and lack context on when/why to use the tool. Input schemas are present but trivial (single string param). No output schema documented. The server appears designed to test MCP infrastructure, not provide real agent capabilities. Duplicate tool names create ambiguity for LLM tool selection.
Returns a demo value. Do not publish sk-demo-secret-123456789.
Echo the input text back.
Echo the input text back.
Returns a demo value. Do not publish sk-demo-secret-123456789.
Duplicate tool names: all 4 tools named 'echo' across different fixtures. LLMs cannot disambiguate which echo to call when multiple are available.
Tool name 'echo' lacks verb_noun structure. Should be 'echo_text' or similar to clarify the action and object.
Descriptions are too short (10-60 chars, baseline 194 chars). 'Echo the input text back.' lacks context on when/why to use this tool or what it returns.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
No output schema documented. LLMs cannot infer what fields or structure the tool returns.
Secret embedded in tool description ('sk-demo-secret-123456789'). Secrets must never appear in descriptions, parameters, or responses, they leak into logs and LLM context.