AI workflows increasingly span multiple products. Legal professionals experience this constantly as they move between drafting, research, document review, and other purpose-built AI experiences while continuing the same task.
That means manually navigating between products, rebuilding context, finding the right materials, and bringing results back into the original workflow. Model Context Protocol (MCP) and similar standards enable AI systems to connect with tools and services while the user stays in the first system. They do not, however, address the user-directed transition between AI experiences, where the context of the work—including the objective, source materials, conversation history, and prior analysis—must move with the user. As a result, valuable context is often lost along the way.
Agent Handoff Protocol eliminates that challenge by enabling that transfer.
What is Agent Handoff Protocol?
Agent Handoff Protocol is an open protocol designed to enable context continuity across AI platforms. It allows users and their work to move between AI experiences with the relevant background information and supporting materials already in place.
The same principle already exists in legal work. When a matter requires specialist expertise, the specialist needs the relevant context to contribute effectively. A tax or environmental lawyer, for example, receives the relevant documents, deal background, and client objectives so they can understand the issue and provide advice or work product the deal team can use.
Why it matters: The work moves, not just the data
With Agent Handoff Protocol, the work can continue even as the AI experience changes.
Users stay in control: They can review, refine, select, and steer the work directly when their judgment is required. Agent Handoff Protocol supports decision-making and keeps users actively involved at the points where their input matters.
The right tool for the task: People use different tools for different kinds of work. Agent Handoff Protocol enables them to move between AI platforms while preserving the work needed to continue the task.
One workflow across AI experiences: Instead of starting over each time they switch platforms, users can continue the same task across products.
The result is less time managing transitions between systems and more time focused on legal analysis, strategy, and client work. It feels like a single conversation rather than a series of disconnected interactions.
How it works
A handoff becomes valuable when the next step calls for a purpose-built experience, direct user involvement, or deeper analysis and interaction than a background connection can provide. Agent Handoff Protocol does not replace MCP or Agent2Agent; each supports a different part of how AI systems work together.
1. An AI agent recognizes that another experience can best serve the next step
The host AI agent recognizes that a different product and interface would be more effective for the next stage of the workflow and offers a seamless continuation.
2. The agent hands off the relevant context – and the user
The user enters the next platform with the objective, materials, conversation history, and thread already in place.
No re-prompting. No re-uploading. No recreating the task.
3. Work in the right interface
The user can inspect, refine, compare, analyze, and steer the work using the tools best suited to the task.
4. Continue the workflow
Selected outputs, supporting materials, and context return to the original workflow so work can proceed as a continuous experience.
Agent Handoff Protocol in Practice
DeepJudge and our integration partners have already begun putting the protocol to work , enabling users to move between specialized AI experiences while continuing the same task.
DeepJudge and Harvey: Powering AI Agents with Institutional Intelligence
Consider a transactional lawyer working in Harvey on a drafting task who needs to find, analyze, and apply clauses from similar prior matters.
When deeper analysis is required—such as identifying the examples most relevant to the transaction at hand, understanding how similar issues were addressed across prior matters, analyzing precedent, or evaluating negotiation positions—the lawyer can move seamlessly between DeepJudge and Harvey with the necessary context already in place.
Learn more about the DeepJudge + Harvey integration
DeepJudge and Thomson Reuters: Grounding Legal Research in Institutional Knowledge
A litigator preparing a motion who wants to build on how similar issues were handled in prior matters may begin in DeepJudge to identify relevant precedent, understand how similar arguments were developed, and surface the most useful examples from prior matters.
From there, the lawyer can continue the task in Thomson Reuters CoCounsel Legal to conduct additional research, refine the analysis, and further develop the work product.
Learn more about the DeepJudge + Thomson Reuters CoCounsel Legal integration
Get Involved
Agent Handoff Protocol is available as an open protocol. DeepJudge welcomes participation from AI providers, specialist platforms, and organizations building AI solutions in-house.
Explore the specification and implementation details on GitHub.
See how DeepJudge works across your AI ecosystem
See how Agent Handoff Protocol enables legal teams to move between AI experiences without rebuilding the task.