Loading ToolBench
AI agent management platform with workspace collaboration, knowledge base search, memory management, web search, MCP server integration, scheduling, eval judges, and webhook/widget deployment
Helpmaton has 5 tools with mixed quality. Naming is generally good (verb-noun pattern followed), but descriptions vary widely in clarity and actionability. Schemas are present but lack critical depth, parameter descriptions are minimal or missing context. The `search` tool has a well-structured enum for category, but most other tools have sparse parameter documentation. No output schemas are documented, and error handling guidance is absent. The server appears to be a Lambda-based backend (not MCP protocol-native), which limits assessment since tool definitions are inferred from TypeScript source rather than declared via MCP protocol.
Get current date and time
Get the current configuration of this agent: name, system prompt, model, enabled tools (document search, memory, web search, etc.), delegatable agents, and memory settings.
Read internal documentation by document ID
Search the web using Exa.ai with category-specific search. This tool allows you to search for specific types of content (companies, research papers, news, PDFs, GitHub repos, tweets, personal sites, people, or financial reports). Use this when you need to find specialized content that matches a specific category.
Update this agent's configuration. Pass only the fields you want to change (e.g. name, systemPrompt, modelName, temperature, or feature flags like enableSearchDocuments).
No output schemas documented for ANY tool. LLMs cannot plan downstream tool calls or extract required fields without knowing response structure.
Parameter descriptions are sparse and lack context. 'Temperature 0-2' omits explanation of what it controls. 'Document ID to read' doesn't specify format, length, or valid characters. LLMs cannot reliably invoke tools with ambiguous parameter guidance.
No error handling or recovery guidance. Tools do not describe error conditions, retryability, or how to recover from failures. E.g., what happens if a docId doesn't exist? Can search fail due to API rate limits?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 52 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 58 | - | v1 |
update_my_config is a write operation but has no confirmation/dry-run pattern. Agents could accidentally modify agent configuration (system prompt, model, tools) without safeguards. Should support a preview or require explicit confirmation.
Tool definitions appear to be inferred from TypeScript source code rather than declared via MCP protocol. Cannot verify actual tool registration, schema validation, or protocol compliance without seeing MCP server code (e.g., tool/list response, Tool interface definitions).
search tool returns results from Exa.ai but no output schema provided. Unknown: how many results? what fields per result? pagination? This forces LLMs to guess at response structure.