Loading ToolBench
Lightweight MCP-native agent framework — connect MCP servers, aggregate tools, run LLM agent loops
Two tools with minimal but present schemas and descriptions. Both tools have clear verb-noun naming (echo, add) and basic input schemas with typed parameters. However, descriptions are extremely brief (10-20 chars), falling below the 34-char p10 baseline for production tools. No output schema documentation. No error handling guidance. No parameter constraints (e.g., numeric ranges for 'add'). Tools are from a mock test fixture, not a production server, limiting real-world applicability.
Add two numbers
Echo back the input text
Descriptions are critically short (10-20 chars). 'Echo back the input text' and 'Add two numbers' lack context for LLM tool selection. Baseline is 34-392 chars; these fall at the p10 floor.
No output schema documented. LLMs cannot plan downstream calls or extract structured data from results. Both tools return text content but this is not declared in the tool definition.
No parameter constraints. 'add' accepts two numbers with no min/max bounds, allowing LLMs to pass absurd values (e.g., 1e308). Numeric parameters should declare ranges.
No error handling guidance. If 'add' receives non-numeric input or 'echo' receives null, there is no recovery hint for the LLM. Error responses should guide next steps.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | <=2025-11-25 | v2 |
Tools are defined in a test fixture (tests/fixtures/mock-server.ts), not in the main server implementation. This is a mock for testing, not a production MCP server.