HumanMCP: A Human-Like Query Dataset for Evaluating MCP Tool Retrieval Performance

TL;DR

This paper presents HumanMCP, a large-scale, human-like query dataset for MCP tools, designed to improve the evaluation of tool retrieval systems by capturing diverse, realistic user interactions.

Contribution

It introduces the first dataset with diverse, high-quality user queries and personas tailored to 2800 tools, addressing gaps in existing benchmarks.

Findings

Enhanced evaluation of MCP tool retrieval performance.

Better representation of real-world user interactions.

Improved generalization in MCP system assessments.

Abstract

Model Context Protocol (MCP) servers contain a collection of thousands of open-source standardized tools, linking LLMs to external systems; however, existing datasets and benchmarks lack realistic, human-like user queries, remaining a critical gap in evaluating the tool usage and ecosystems of MCP servers. Existing datasets often do contain tool descriptions but fail to represent how different users portray their requests, leading to poor generalization and inflated reliability of certain benchmarks. This paper introduces the first large-scale MCP dataset featuring diverse, high-quality diverse user queries generated specifically to match 2800 tools across 308 MCP servers, developing on the MCP Zero dataset. Each tool is paired with multiple unique user personas that we have generated, to capture varying levels of user intent ranging from precise task requests, and ambiguous, exploratory commands, reflecting the complexity of real-world interaction patterns.