Agent Context Protocols Enhance Collective Inference

TL;DR

This paper introduces Agent Context Protocols (ACPs), structured communication protocols that improve multi-agent collaboration, leading to state-of-the-art performance in complex AI tasks like web assistance and multimodal report generation.

Contribution

The paper presents ACPs, a novel family of structured, domain- and agent-agnostic protocols that enhance coordination, robustness, and extensibility in multi-agent AI systems.

Findings

Achieved 28.3% accuracy on AssistantBench for web assistance

Outperformed commercial AI systems in human evaluations

Enabled rapid development of high-performing generalist agents

Abstract

AI agents have become increasingly adept at complex tasks such as coding, reasoning, and multimodal understanding. However, building generalist systems requires moving beyond individual agents to collective inference -- a paradigm where multi-agent systems with diverse, task-specialized agents complement one another through structured communication and collaboration. Today, coordination is usually handled with imprecise, ad-hoc natural language, which limits complex interaction and hinders interoperability with domain-specific agents. We introduce Agent context protocols (ACPs): a domain- and agent-agnostic family of structured protocols for agent-agent communication, coordination, and error handling. ACPs combine (i) persistent execution blueprints -- explicit dependency graphs that store intermediate agent outputs -- with (ii) standardized message schemas, enabling robust and fault-tolerant multi-agent collective inference. ACP-powered generalist systems reach state-of-the-art performance: 28.3 % accuracy on AssistantBench for long-horizon web assistance and best-in-class multimodal technical reports, outperforming commercial AI systems in human evaluation. ACPs are highly modular and extensible, allowing practitioners to build top-tier generalist agents quickly.