Loading ToolBench
This is a CLI tool and test fixture suite rather than a production MCP server. The 12 tools included are test fixtures spread across mock servers (mock-server.ts, mock-http-server.ts, create-v2-mock-server.ts, mock-server-2.ts). While tool definitions ARE visible in source code and most have descriptions and schemas, the quality is uneven. Tool names are reasonable (echo, add, multiply, delete_world, read_status) but some lack verb clarity. Descriptions range from adequate ("Echoes back the input") to vague ("A secret tool"). Parameter descriptions are present but often minimal. Output schemas are mentioned in fixture code but not explicitly documented. Critical gaps: no error handling patterns are visible, no validation rules documented, and several tools have placeholder or minimal descriptions. This server is designed as a testing harness, not production tooling, which explains the lower quality baseline.
Adds two numbers
Task-augmented echo that pauses in input_required
Asks for user confirmation via elicitation
An open-world writeable tool (annotated)
Echoes back the input
Task-augmented call that never replies (timeout tests)
Multiplies two numbers
Tool 'secret' has minimal description ('A secret tool'), fewer than 20 characters, providing no context for when/why to call it. Description score capped at 0-20 per hard scoring rules.
No output schemas are documented for any tool. While input schemas are present and valid, LLMs cannot plan downstream tool calls without knowing what fields and types to expect in responses. Mock servers return hardcoded responses (e.g., 'echo' returns {content: [{type: 'text', text: ...}]}) but this structure is not declared in tool metadata.
Parameter descriptions are minimal and generic. For example, 'echo' and 'slow_echo' both describe message as 'Message to echo', lacking context about format, length limits, or constraints.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 66 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 41 | 2025-03-26+ | v1 |
A tool that takes no arguments
A read-only tool (annotated)
A secret tool
Echoes back the input after a simulated delay (supports tasks)
Returns a marker proving a modern MCP tool call completed
No error handling patterns visible. Tools do not document what happens on failure, how to recover, or whether calls are retryable. Per pattern:recovery-guide, error responses should guide the agent on next steps.
Tool 'delete_world' is marked DESTRUCTIVE but lacks a confirmation or dry-run pattern. Per pattern:confirmation-request, irreversible operations should support confirmation or dry-run to prevent agent mistakes.
Several tool names lack clear action verbs. 'noop', 'secret', and 'hang_task' do not signal what action occurs. Per naming baseline, 90% of A+ tools start with action verbs (get, list, create, search, update, delete, send).