AWCP: A Workspace Delegation Protocol for Deep-Engagement Collaboration across Remote Agents

TL;DR

The Agent Workspace Collaboration Protocol (AWCP) is introduced, which bridges the gap through temporary workspace delegation inspired by the Unix philosophy that everything is a file, and paves the way for a universally interoperable agent ecosystem in which collaboration transcends message boundaries.

Abstract

The rapid evolution of Large Language Model (LLM)-based autonomous agents is reshaping the digital landscape toward an emerging Agentic Web, where increasingly specialized agents must collaborate to accomplish complex tasks. However, existing collaboration paradigms are constrained to message passing, leaving execution environments as isolated silos. This creates a context gap: agents cannot directly manipulate files or invoke tools in a peer's environment, and must instead resort to costly, error-prone environment reconstruction. We introduce the Agent Workspace Collaboration Protocol (AWCP), which bridges this gap through temporary workspace delegation inspired by the Unix philosophy that everything is a file. AWCP decouples a lightweight control plane from pluggable transport mechanisms, allowing a Delegator to project its workspace to a remote Executor, who then operates on the shared files directly with unmodified local toolchains. We provide a fully open-source reference implementation with MCP tool integration and validate the protocol through live demonstrations of asymmetric collaboration, where agents with complementary capabilities cooperate through delegated workspaces. By establishing the missing workspace layer in the agentic protocol stack, AWCP paves the way for a universally interoperable agent ecosystem in which collaboration transcends message boundaries. The protocol and reference implementation are publicly available at https://github.com/SII-Holos/awcp.

Figures (7)

• Figure 1: Two paradigms for inter-agent collaboration. (a) In message-based protocols such as A2A and ANP, agents exchange serialized task descriptions and results; the receiving agent operates on detached artifacts without access to the originator's execution environment, leading to well-documented failure modes including context loss, incomplete information transfer, and environment mismatch cemri2025mast. (b) In AWCP's deep-engagement model, the Delegator projects its workspace to the Executor, who operates directly on the original files using its local toolchain, preserving full context, providing the complete environment, and enabling in-situ execution.
• Figure 2: AWCP in the agentic protocol landscape. Delegator agents access AWCP through integration interfaces such as MCP tool servers or skill modules. The control plane uses standard HTTP and Server-Sent Events for signaling between AWCP services. Executor agents are invoked through an extensible adapter layer supporting A2A, ANP, or direct HTTP. The transport plane provides pluggable filesystem-level access that complements message-level coordination.
• Figure 3: AWCP architecture. The control layer handles signaling and state synchronization between the Delegator and Executor, while the transport layer provides pluggable adapters (SSHFS, Archive, Storage, Git) for data exchange. Both nodes run dedicated AWCP services while agents operate with their native toolchains.
• Figure 4: AWCP four-phase message sequence between Delegator and Executor. Solid, dashed, and blue arrows denote synchronous HTTP requests, responses, and asynchronous SSE events, respectively.
• Figure 5: Dual state machines governing the AWCP delegation lifecycle. The DelegationStateMachine (left) tracks nine states on the Delegator side; the AssignmentStateMachine (right) tracks four states on the Executor side. Solid arrows trace the happy path; a single dashed red path indicates that any non-terminal state may transition to a terminal state upon error, cancellation, or lease expiration. Blue dashed arrows mark cross-machine synchronization via protocol messages.