Title: MCP-Persona: Evaluating LLM Agents on Real-World Personal Applications Through Environment Simulation

Abstract:

The Model Context Protocol (MCP) has become a pivotal standard for linking large language models (LLMs) with external tools and data sources, seeing swift integration into development platforms and personal software. Despite this rapid adoption, current evaluation benchmarks largely concentrate on general-purpose information retrieval tools, neglecting the unique complexities of personal social applications. In these contexts, tools must interact with specific user accounts or local databases, presenting distinct challenges. To address this oversight, we present MCP-Persona, the inaugural benchmark tailored to assess agent efficacy with real-world, customized MCP tools. This benchmark covers a broad spectrum of popular applications, including social networks such as Reddit and Xiaohongshu (Rednote), as well as enterprise communication platforms like Slack and Lark (Feishu). Our comprehensive testing of several state-of-the-art (SOTA) agents reveals substantial difficulties in utilizing personalized tools, underscoring the benchmark’s importance in uncovering and resolving these performance gaps. MCP-Persona is openly accessible at https://github.com/wwh0411/MCP-Persona.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC