metadata
license: mit
task_categories:
- text-generation
- question-answering
tags:
- mcp
- ai-agents
- benchmark
- evaluation
- tool-use
- function-calling
size_categories:
- n<1K
Agent Evaluation Benchmark
A benchmark dataset for evaluating AI agent tool-use capabilities across 55+ test cases spanning 14 categories.
Overview
This benchmark tests whether AI agents can correctly select and use the right MCP tools for real-world tasks. It covers data retrieval, blockchain queries, security analysis, academic research, and more.
Categories
| Category | Test Cases | Description |
|---|---|---|
| Weather | 5 | Forecasts, UV index, climate history |
| Blockchain | 7 | Token prices, wallet analysis, DeFi, whale tracking |
| Security | 5 | CVE search, vulnerability analysis, CVSS scores |
| Academic | 4 | Paper search, citations, author lookup |
| Company | 3 | EU company registry search |
| Agriculture | 3 | Crop data, food prices, yield comparison |
| Space | 4 | NASA APOD, asteroids, Mars rover, ISS |
| Aviation | 3 | Flight tracking, airport info |
| Medical | 3 | WHO data, disease outbreaks, health stats |
| Political | 2 | Campaign finance, FEC data |
| Supply Chain | 2 | UN trade data, import/export stats |
| LLM Benchmark | 3 | Model comparison, pricing, benchmarks |
| Energy | 3 | CO2 intensity, energy mix, electricity prices |
| Legal | 3 | Court decisions, case search |
| Agent Infrastructure | 3 | Directory, memory, workflows |
| Compliance | 2 | PII detection, GDPR checks |
Schema
- task_description: Natural language description of the task
- expected_tool: The MCP tool that should be selected
- difficulty: easy, medium, or hard
- category: Task category
- test_input: Example input parameters
- expected_output_contains: Key string that should appear in the output
Difficulty Distribution
- Easy: 25 tasks (basic single-tool queries)
- Medium: 25 tasks (parameter selection, filtering, comparison)
- Hard: 5 tasks (multi-step reasoning, complex analysis)
Usage
Use this dataset to evaluate:
- Tool Selection Accuracy: Does the agent pick the right tool?
- Parameter Extraction: Does the agent correctly parse inputs?
- Output Validation: Does the response contain expected information?
from datasets import load_dataset
ds = load_dataset("aiagentkarl/agent-evaluation-benchmark")
License
MIT