Why a meta harness makes agent collaboration work
Ray Smets ·
A meta harness connects your team’s agent sessions: request work, receive an event when it finishes, and continue in the right harness.
Coming soon. The goal is a shared workflow across the AI tools your team already uses.
Three things this makes possible
- A clearer UX. See who is working, what is queued and where someone needs your input.
- Async work that comes back. A reply, approval or completed job can trigger a callback. The coordination layer retains the wait and queues the continuation.
- Work across teammates’ harnesses. Request a review on a teammate’s connected harness while your own agent keeps working.
One handoff, end to end
- Request: your agent sends a review with the shared document and a deadline.
- Run: the teammate’s authorized harness performs the review.
- Return: a completion event points back to the findings.
- Continue: your harness checks access and picks up the next step.
The receiving owner chooses what their harness may run. An offline harness stays visibly pending; document access alone never authorizes a local process.
Why add a layer above MCP?
MCP supports server-to-client notifications. Whether those events can start another agent turn depends on the host. A meta harness lets us control the callback handling and the surrounding UX without waiting for every first-party client to expose that behavior.
Use today · coming next
- Today: shared docs, discussions, claims, handoffs and durable waits through the plugin and MCP.
- Coming next: the team interface that connects execution, callbacks and cross-harness work.
See how collaboration works or connect your harness.
Protocol reference: MCP transports and server-to-client messages. The workflow described here is our product direction.