Loading ToolBench
A collection of AI agent implementations demonstrating various patterns including sequential agents, parallel agents, loop-based agents, and tool integration with Google ADK and MCP
This repository is a multi-agent tutorial/demo, not a production MCP server. Tool definitions exist but lack rigor: descriptions are present but minimal (10-50 chars), parameter descriptions are sparse or missing, no output schemas documented, no error handling guidance, and no input validation. The 4 tools show basic structure but fall far short of production standards. Transport is unknown (appears to be in-memory/demo only, not MCP-compliant).
Get product information for a given product.
Tool to retrieve user name and country from session state.
Tool to record and save user name and country in session state.
Sets the status of a smart home device.
No output schemas documented. Tools return dicts/strings but LLMs cannot infer field structure, types, or downstream chaining requirements.
Parameter descriptions are missing or trivial. 'status' in set_device_status says 'either ON or OFF' but lacks context on what happens if invalid input is passed, or whether the tool is idempotent.
No error handling or recovery guidance. Tools return success/failure dicts but do not explain what to do on failure, whether errors are retryable, or what the LLM should try next.
Tool descriptions are too short (10-50 chars). Baseline for A+ tools is 194 chars. Descriptions lack WHEN to use, WHAT it returns, and dependencies.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 37 | <=2025-11-25 | v2 |
No input validation or constraint documentation. set_device_status accepts 'status' as free-form string; should be enum (ON|OFF). No min/max for numeric params, no regex patterns, no length limits.