Overview
OpenWOP, short for Workflow Orchestration Protocol, defines how multi-agent workflows start, stream events, pause for human approval, resume, and replay. It uses REST and Server-Sent Events so clients can work with any compatible host rather than depending on one vendor's runtime.
The protocol is intended for developers and teams building agent applications that take consequential actions such as making payments, sending messages, or changing records. It standardizes the workflow run while leaving agent and tool connectivity to complementary technologies.
Key Capabilities
- Human approval can stop a run before a sensitive action and resume it from saved state after a decision.
- Replay can fork a completed run from an earlier step, reuse recorded outcomes, and suppress repeated side effects.
- Portable exports can move agents, packs, schedules, and workflows between compatible hosts while secrets remain at the original host.
- A shared event vocabulary records decisions, tool calls, and handoffs, with OpenTelemetry tracing for cross-host inspection.
- Run limits can cap steps, elapsed time, and spending, with budget reserved before execution.
Interoperability
OpenWOP is a protocol rather than an agent framework or workflow engine. A compatible TypeScript or Python host can execute the same workflow definition with matching final state and event-log structure for core workflows. Host capability discovery lets clients negotiate supported behavior.
It does not replace MCP or A2A. An OpenWOP host can call MCP tools and A2A agents inside a run while treating their returned data as untrusted input.
Development Resources
The specification, JSON schemas, and API definitions are licensed under CC BY 4.0. The Apache 2.0 code includes TypeScript, Python, and Go SDKs, a command-line interface, a conformance suite, and published packs.
Developers can begin with the hosted reference app or run the reference host locally with Node.js. The documented quickstart covers an initial run, approvals, replay, and webhooks.
Suitable Use Cases
OpenWOP is designed for multi-agent processes that need oversight, traceability, and portability. Supported scenarios include finance approvals, audit replay, long-running research, cross-host customer support, voice ordering with declarative forms, and moving workflows between cloud hosts.
For a single-agent application, a dedicated agent framework may add less overhead. The project also notes that mature durable execution or visual automation platforms may be a better fit when those capabilities are the primary requirement.