MCP Agent Trajectory Benchmark with 38 Distinct Business Domains

38 distinct MCP agents provide complete tool-use trajectories in the ATIF v1.2 format. Each agent operates in a distinct business domain with custom tools, realistic user conversations, and full execution traces. The dataset is designed for training and evaluating tool-use and function-calling capabilities of LLMs.

Use Cases

• Train LLMs on agent trajectories to improve function-calling accuracy across distinct business domains.
• Benchmark the tool-use performance of language models using realistic user conversations and full execution traces.
• Analyze the structure and patterns within ATIF v1.2 formatted trajectories for agent behavior research.

Strengths

• Contains 38 distinct agent roles, each with a unique operational domain.
• Provides complete tool-use trajectories, including realistic user conversations and execution traces.
• Uses the standardized ATIF v1.2 format for agent trajectory interchange.

Limitations

• The dataset size, number of rows, and specific column schema are unknown.
• Potential domain bias is introduced as the 38 agents cover distinct but unspecified business areas.
• The absence of sample data and file format details limits immediate usability assessment.

Provenance

Source

huggingface, author obaydata

Collection Method

Benchmark dataset of MCP agent trajectories.

Freshness

Last updated on 2026-03-19.