Loading ToolBench
An MCP server for testing and inspecting MCP client implementations. Provides tools and console integration for testing sampling, elicitation, and notification handling.
This MCP server presents a test/inspector-focused tool suite with critically insufficient documentation. All 6 tools have minimal descriptions (10-50 chars) that are far below the 194-char baseline and fail to explain WHAT the tool does, WHEN to use it, or what it returns. No input schemas are visible in the provided source code, tool definitions appear to be registered but their parameter structures, types, and constraints are completely absent. The code references 'mcp-server/main.ts' but that file is not provided, preventing verification of actual schema definitions. Without schemas and detailed descriptions, these tools cannot be reliably selected or called by LLMs. The naming is adequate (verb_noun style: echo, convert_date, calculate) but generic descriptions like 'Inspector tool for echoing input' are non-descriptive and do not guide agent behavior.
Inspector tool for calculations
Inspector tool for date conversion
Inspector tool that delays response
Inspector tool for echoing input
Inspector tool for generating random data
Inspector tool for triggering error responses
No input schemas visible for any tool. Tool definitions are referenced in mcp-server/main.ts but that file was not provided. Per HARD SCORING RULES, if a tool has NO input schema, schema score MUST be 0.
All tool descriptions are under 20 characters (e.g., 'Inspector tool for echoing input' = 34 chars but non-descriptive). Per HARD SCORING RULES, descriptions under 20 chars cannot exceed 40, and these generic inspector descriptions fail to explain what the tool does, when to use it, or what it returns. Baseline is 194 chars.
Parameter descriptions are completely missing. No evidence of any parameter-level documentation. LLMs cannot infer parameter meaning from names alone, a 'status' param could mean HTTP status, project status, or boolean active flag. Baseline: 100% of A+ tools have descriptions on every parameter.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 34 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 46 | - | v1 |
No output schema documentation. LLMs need to know what fields to expect from these tools so they can plan downstream calls and extract the right data. No evidence of return type documentation in the provided source.
No error handling guidance. Tools like 'trigger_error' suggest testing of error paths, but no documentation explains how errors are classified, whether they are retryable, or what recovery actions an agent should take.
Naming is generic and does not follow the verb_noun pattern consistently. 'echo', 'calculate', 'random_data', and 'trigger_error' are vague about their purpose. 'echo' could mean echo input to output, or echo a message, the description does not clarify. 'calculate' with no context on what is calculated.