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

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Laddha, Shubh, Changbencharoen, Lucas, Kuptivej, Win, Shringla, Surya, Vaidheeswaran, Archana, Bhaskar, Yash
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911472807837696
author Laddha, Shubh
Changbencharoen, Lucas
Kuptivej, Win
Shringla, Surya
Vaidheeswaran, Archana
Bhaskar, Yash
author_facet Laddha, Shubh
Changbencharoen, Lucas
Kuptivej, Win
Shringla, Surya
Vaidheeswaran, Archana
Bhaskar, Yash
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2602_23367
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle HumanMCP: A Human-Like Query Dataset for Evaluating MCP Tool Retrieval Performance
Laddha, Shubh
Changbencharoen, Lucas
Kuptivej, Win
Shringla, Surya
Vaidheeswaran, Archana
Bhaskar, Yash
Artificial Intelligence
Information Retrieval
68T01, 68T50
I.2.11; H.3.3; I.2.7
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.
title HumanMCP: A Human-Like Query Dataset for Evaluating MCP Tool Retrieval Performance
topic Artificial Intelligence
Information Retrieval
68T01, 68T50
I.2.11; H.3.3; I.2.7
url https://arxiv.org/abs/2602.23367