Loading ToolBench
An MCP server for Agent5ive, built with the official @modelcontextprotocol/sdk.
This server has 2 tools with significant definition gaps. Both tools have descriptions, but they are generic and lack detail. The first tool (get_agent_purpose) has an empty input schema ({}), which violates the requirement for parameter documentation. The second tool (interact_with_agent) uses Zod schema definitions that are visible in code but not explicitly documented in parameter descriptions within the schema itself. Error handling exists but returns raw error messages without recovery guidance. No output schemas are documented. Neither tool follows the chat data model pattern, they require a DEPLOYMENT_ID environment variable rather than accepting it as a user-friendly parameter or using natural identifiers.
Get the details and purpose of the deployed agent.
Interact with the deployed agent.
get_agent_purpose has an empty input schema ({}), providing no documentation of parameters or expected behavior.
Descriptions are too generic and lack actionable context. 'Get the details and purpose of the deployed agent' (47 chars) does not explain WHEN to call it, what it returns, or how it differs from interact_with_agent. These are below the lower percentile.
No output schemas documented. LLMs cannot determine what fields to expect in responses. LLMs need to know what fields to expect so they can plan downstream tool calls and extract the right data.' Both tools return raw JSON or plain text without structured definitions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 11 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 45 | - | v1 |
DEPLOYMENT_ID is injected via environment variable or URL query parameter, not as a tool parameter. This violates the natural-identifiers pattern. Users say 'interact with my agent' but the tool requires server configuration rather than user-facing input. Accept usernames, emails, and display names, resolve them to system IDs inside the tool.'
Error messages are generic and do not guide recovery. Examples: 'Error: DEPLOYMENT_ID is not set.' and 'Error: <error.message>'. Try search_users() with a partial name." A raw error code or stack trace gives the agent nothing to act on.'
interact_with_agent's history parameter has optional array of objects with role/content fields, but parameter descriptions in the schema lack detail. The description 'The conversation history.' is too brief. LLMs cannot infer parameter meaning from names alone.'
No error classification or recovery hints provided. An LLM should know: can I retry? Should I ask the user? Or is this unrecoverable?' Current error handling returns isError=true but provides no guidance on whether the error is transient.
Tool composition is incomplete. interact_with_agent requires pre-existing conversation history as input, but there is no tool to retrieve conversation history or session state. Agents must manually construct history, violating the composition pattern.