[Paper Review] Human Tool: An MCP-Style Framework for Human-Agent Collaboration

Yuanrong Tang, Peng, Huiling+6 more|arXiv (Cornell University)|Feb 13, 2026

Human-Automation Interaction and Safety0 citations

TL;DR

The paper proposes Human Tool, an MCP-style abstraction that treats humans as callable tools within AI-led workflows, improving performance and reducing workload in AI-advantaged tasks. Controlled studies show better outcomes and more balanced collaboration than AI-only baselines.

ABSTRACT

Human-AI collaboration faces growing challenges as AI systems increasingly outperform humans on complex tasks, while humans remain responsible for orchestration, validation, and decision oversight. To address this imbalance, we introduce Human Tool, an MCP-style interface abstraction, building on recent Model Context Protocol designs, that exposes humans as callable tools within AI-led, proactive workflows. Here, "tool" denotes a coordination abstraction, not a reduction of human authority or responsibility. Building on LLM-based agent architectures, we operationalize Human Tool by modeling human contributions through structured tool schemas of capabilities, information, and authority. These schemas enable agents to dynamically invoke human input based on relative strengths and reintegrate it through efficient, natural interaction protocols. We validate the framework through controlled studies in both decision-making and creative tasks, demonstrating improved task performance, reduced human workload, and more balanced collaboration dynamics compared to baseline systems. Finally, we discuss implications for human-centered AI design, highlighting how MCP-style human tools enable strong AI leadership while amplifying uniquely human strengths.

Motivation & Objective

• Motivate the need to shift orchestration from humans to AI in AI-advantaged tasks to reduce coordination bottlenecks.
• Define a structured Human Tool abstraction with capabilities, information, and authority for AI to call human input.
• Implement Human Tool as an MCP-style interface with tool schemas and invocation protocols.
• Demonstrate through controlled experiments that Human Tool improves task performance and lowers perceived workload.

Proposed method

• Define Human Tool as a structured, callable abstraction representing a human contributor along three dimensions: Capabilities, Information, and Authority.
• Determine when to invoke Human Tool through hierarchical task analysis and three invocation conditions: capability complementarity, information exchange, and authority control.
• Establish communication patterns with humans using interaction behaviors and lightweight guidelines to minimize coordination overhead.
• Implement the framework as an MCP-style interface with a Python backend (LangGraph), GPT-4o reasoning, MySQL storage, and a React TypeScript frontend for structured orchestration.
• Conduct controlled experiments comparing Human Tool to an AI Tool baseline across two task domains (Travel Planning and Story Writing) using standardized usability and workload measures and objective task outcomes.

Figure 1 : Contrasting paradigms of human-AI collaboration: Human Tool versus AI Tool. In the Human Tool paradigm, humans are exposed to the agent as MCP-style callable interfaces rather than workflow leaders.

Experimental results

Research questions

• RQ1 RQ1: Does Human Tool improve task performance and reduce workload compared to an AI tool baseline?
• RQ2 RQ2: Is the framework adaptable across different task types (planning vs. creative tasks)?
• RQ3 RQ3: Does Human Tool enhance engagement and collaboration dynamics in human–AI interaction?

Key findings

• Human Tool outperformed the AI Tool baseline on both tasks, with higher task accuracy (Travel Planning: 86.72% vs 72.66%; Story Writing: 68.38 vs 58.56) and higher winner rates (Story Writing: 0.611 vs 0.371).
• Participants reported lower cognitive load and higher collaboration satisfaction under Human Tool, with CSI means of 75.48 vs 52.83 and significant reductions in mental effort for Story Writing.
• Usability (SUS) was higher for Human Tool in both tasks (Travel Planning: 70.89 vs 58.22; Story Writing: 79.79 vs 60.21).
• Invocation logs show human input was requested at decision-critical points, improving input relevance and timing (preferences, responsibility boundaries, knowledge identification).
• Qualitative interviews indicated participants viewed the system as a partner, enabling deeper joint exploration and smoother co-development of ideas.

Figure 2 : An integrated framework for representing Human Tool within AI-managed workflows. It organizes three dimensions: How to define Human Tool, when to call them, and how to communicate effectively.