Loading ToolBench
Inspect, call, smoke-test and document any Model Context Protocol server from the command line.
This is a test/demo server with minimal production intent. All 5 tools have basic schemas and most have descriptions, but quality is inconsistent. Descriptions are terse (10-50 chars, below the 194-char baseline). Parameters lack descriptions entirely, LLMs cannot infer what 'a', 'b', 'x', 'seconds' mean without context. No output schemas documented. No error handling guidance. Tool names are generic (add, echo, fail, slow) rather than action-verb patterns. This server is fit for testing the mcpcurl CLI tool itself, not for production agent use.
Add two integers.
Return the text, optionally upper-cased.
Always raises, to exercise error reporting.
Sleep for a while, to exercise latency budgets.
All parameters lack descriptions. LLMs cannot infer meaning of 'a', 'b', 'x', 'seconds', 'text', 'upper', 'message'. Parameter descriptions are required for every input.
No output schemas documented. LLMs do not know what fields to expect from tool responses, blocking downstream composition and data extraction.
Tool names are generic (add, echo, fail, slow) rather than action-verb patterns. 'fail' and 'slow' do not convey intent; 'add' lacks context (add what?). Names should be verb_noun (e.g., compute_sum, format_text).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 40 | 2025-06-18+ | v2 |
No error handling guidance. The 'fail' tool raises RuntimeError but provides no recovery hints. Error responses must tell LLMs what to do next.