ACPX: Cross-Platform Agent Communication Protocol

Something fundamental is broken in the AI agent ecosystem of 2026. Hundreds of agent frameworks exist – OpenClaw, LangGraph, CrewAI, AutoGPT, Semantic Kernel, and countless others – yet most of them cannot talk to each other. An agent built on LangGraph has no standard way to delegate a task to an agent running on OpenClaw. A CrewAI swarm cannot discover or invoke a specialist agent running on a different platform. This fragmentation is holding back the entire field from realizing the vision of a genuinely interoperable agent internet.

ACPX (Agent Communication Protocol X) is OpenClaw’s answer to this problem. It is an open, cross-platform protocol that defines how AI agents discover each other, negotiate capabilities, exchange messages, and collaborate across framework boundaries – whether those agents run on a laptop, a VPS, or a hyperscaler’s GPU cluster.

What makes ACPX distinctive is its design philosophy: it assumes nothing about the agent’s internal architecture. Whether your agent is a simple RAG pipeline, a complex multi-agent orchestration system, or a single-model chatbot, ACPX provides a standard envelope for communication. The protocol handles discovery (finding agents with the right capabilities), authentication (verifying identity across trust boundaries), routing (delivering messages to the right agent instance), and serialization (framing agent messages in a language-agnostic format).

Core Architecture

ACPX is built on four foundational layers, each solving a specific interoperability challenge:

Discovery in Detail

The discovery layer is arguably ACPX’s most innovative component. Agents register their capabilities in a decentralized directory using either local network discovery (mDNS for same-machine or same-network agents) or a distributed hash table (DHT) for internet-scale discovery. Each registration includes:

• Agent identity and public key
• Declared capabilities in a machine-readable capability schema
• Supported transport protocols and endpoints
• Trust anchors and attestation proofs

This means an agent running on a developer’s laptop can discover and invoke a GPU-accelerated inference agent running on a cloud cluster, without either side knowing the other’s framework details.

ACPX vs. Other Protocols

The agent communication protocol landscape is growing crowded. Here is how ACPX compares to the other major contenders:

A Multi-Agent Communication Flow

The following Mermaid diagram illustrates how ACPX enables a typical multi-agent workflow where agents on different frameworks collaborate to complete a complex task:

sequenceDiagram
    participant User as User
    participant Orchestrator as Orchestrator Agent<br>(OpenClaw)
    participant Registry as ACPX Registry<br>(DHT)
    participant Researcher as Research Agent<br>(LangGraph)
    participant Coder as Code Agent<br>(AutoGPT)

    User->>Orchestrator: "Research and implement a CSV parser"
    Orchestrator->>Registry: Discover agents with "research" capability
    Registry-->>Orchestrator: Found Research Agent (LangGraph)
    Orchestrator->>Research Agent: ACPX:Request("Find CSV parsing libraries")
    Research Agent-->>Orchestrator: ACPX:Response("Found 3 Python libraries")
    Orchestrator->>Registry: Discover agents with "code-gen" capability
    Registry-->>Orchestrator: Found Code Agent (AutoGPT)
    Orchestrator->>Code Agent: ACPX:Request("Implement CSV parser using library X")
    Code Agent-->>Orchestrator: ACPX:Response(implementation_code)
    Orchestrator->>User: "Task complete. CSV parser ready."

Key Specifications

ACPX defines several key message types that agents use to communicate:

• ACPX:Hello – Initial handshake and capability advertisement
• ACPX:Request – A task or query sent to another agent
• ACPX:Response – The result or answer from a request
• ACPX:Progress – Streaming progress updates during long-running tasks
• ACPX:Delegate – Forward a subtask to another agent with context
• ACPX:Error – Structured error reporting across agent boundaries
• ACPX:Bye – Graceful connection termination

These message types create a universal vocabulary that any agent framework can implement, reducing the integration burden for framework authors and enabling a richly connected agent ecosystem.

Getting Started with ACPX

Developers interested in implementing or integrating with ACPX should start with the ACPX specification on GitHub. The repository includes:

• The full protocol specification in OpenAPI 3.1 format
• Reference implementations in Python, TypeScript, and Go
• Example agents demonstrating cross-platform communication
• Conformance test suites for validating implementations

The OpenClaw documentation portal also includes tutorials for connecting OpenClaw agents to external frameworks via ACPX, with step-by-step guides for LangGraph, CrewAI, and Semantic Kernel integrations.

FAQ

What is ACPX?

ACPX (Agent Communication Protocol X) is an open, cross-platform protocol that enables different AI agent frameworks and systems to communicate, coordinate, and share information with each other, regardless of their underlying architecture or implementation language.

Why is cross-platform agent communication important?

As the AI agent ecosystem has exploded with hundreds of frameworks, tools, and platforms, the lack of interoperability has created silos. ACPX aims to be the universal layer that allows agents built on different frameworks to discover, trust, and collaborate with one another seamlessly.

How does ACPX differ from MCP or A2A?

While MCP (Model Context Protocol) focuses on tool and resource access for individual models, and A2A (Agent-to-Agent) targets agent coordination from Google, ACPX is designed as a general-purpose cross-platform communication layer that supports discovery, negotiation, messaging, and security across heterogeneous agent systems.

What use cases does ACPX enable?

ACPX enables multi-agent swarms spanning different frameworks, federated research across institutional boundaries, cross-platform automation pipelines, marketplace-style agent discovery, and enterprise interoperability where different departments use different agent stacks.

Is ACPX an open standard?

Yes, ACPX is developed as an open standard under OpenClaw’s foundation governance. The specification, reference implementations, and SDKs are available on GitHub under permissive licensing, and the community is encouraged to contribute protocol extensions and language bindings.

Further Reading

• ACPX GitHub Repository – Protocol specification, reference implementations, and SDKs
• OpenClaw Official Site – Documentation, tutorials, and community resources
• MCP Specification by Anthropic – The Model Context Protocol for model-to-tool integration
• OpenClaw Complete Guide 2026 – Comprehensive guide to setting up and using OpenClaw