Loading ToolBench
Measure what AI agents do on your website: referrals, fetches, WebMCP tool calls and conversions. One line of script, no cookies, no personal data.
Single tool 'lookup_score' has a clear verb-noun name and basic description, but critical gaps prevent higher scoring. The description (95 chars) is below the 194-char baseline and lacks WHEN/WHY context. The input schema is present with proper JSON Schema structure, but the 'host' parameter description is minimal (47 chars) and lacks format constraints (e.g., 'must contain a dot, no protocol prefix'). No output schema is documented, the tool returns text but LLMs cannot predict the structure. No error handling guidance (what if host is invalid? what if the scan fails?). The tool is defined in a React component (DemoTools.tsx) rather than a dedicated MCP server implementation, suggesting this is a demo/client-side tool, not a production MCP server tool. This is a significant architectural issue.
Look up the Agent Readiness Score of a public website by hostname.
Tool defined in React component, not MCP server. DemoTools.tsx registers 'lookup_score' client-side via document.modelContext.registerTool(), not via MCP protocol. This is a WebMCP client tool, not an MCP server tool. The server itself has no tool definitions visible.
No output schema documented. Tool returns { content: [{ type: 'text', text: string }] } but LLMs cannot infer this structure. Must document: 'Returns a text response with format: {host}: {score}/100 or {host}: n/a/100 if unavailable.'
Parameter description lacks format constraints. 'host' description is 'Hostname, for example example.com' (47 chars). Must specify: 'A valid hostname without protocol (e.g., example.com, not https://example.com). Must contain at least one dot.'
No error handling guidance. Tool throws TypeError for invalid input but provides no recovery path. Must document: 'If host is invalid, returns error. Ensure hostname contains a dot and no protocol prefix. If scan fails, returns score: n/a.'
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | 2026-07-28+ | v2 |
Tool description lacks WHEN/WHY context. Current: 'Look up the Agent Readiness Score of a public website by hostname.' (95 chars, below 194-char baseline). Should explain: 'Call this to check if a website is ready for agent interactions. Returns a score 0-100 indicating WebMCP tool support, tracking integration, and agent-friendly features.'