Loading ToolBench
Drop-in MCP replacement for built-in WebFetch tools, with support for domain-scoped custom HTTP headers (e.g. provider-specific authentication tokens), retry logic, configurable timeouts, per-domain proxies, and flexible output formats.
Single tool 'fetch' has a well-structured schema with 7 parameters, all typed and described. Description is comprehensive (180+ chars) and explains the tool's purpose, supported formats, and key features. Parameter descriptions are detailed and actionable (e.g., 'CSS selector to extract specific elements from the page'). However, output schema is not documented, LLMs cannot predict the structure of returned data. Error handling guidance is absent from the tool definition. The tool name 'fetch' is generic but acceptable for a single-purpose server. No security-sensitive parameters are exposed (good practice). Overall solid definition quality with room for output schema documentation and error recovery guidance.
Fetch the content of a URL and return it in the specified format (raw, markdown, trafilatura, json, or lighthtml). Supports custom headers, retries, timeouts, proxies, and bot-block detection.
Output schema not documented. Tool description does not specify the structure of returned data (fields, types, format). LLMs cannot plan downstream operations or extract specific fields without knowing the response shape.
No error handling guidance in tool definition. Tool description lacks recovery hints (e.g., 'If bot-block detected, retry with render_js=true' or 'If timeout, increase timeout parameter'). Agents cannot self-correct on failures.
Parameter 'output_format' accepts free-form string despite having known valid values. Should be declared as enum: ['raw', 'markdown', 'trafilatura', 'json', 'lighthtml'] to prevent hallucinated formats.
Numeric parameter 'timeout' lacks bounds. Description says 'Defaults to server config or 30' but does not specify min/max range. LLMs could pass absurd values (0, 999999) causing timeouts or resource exhaustion.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | <=2025-11-25 | v2 |
Parameter 'max_bytes' description mentions '-1 to disable' but does not explain the default (10MB) or why disabling is risky. Agents may not understand the memory implications of passing -1.