MCP

MCP is already a control surface.

The CLI owns execution. MCP exposes those runs to clients that need status, event summaries, evidence, diffs, handoff decisions and resume control. If a team later changes its client or agent product, the run semantics should not move with it.

AVAILABLE Route status

Control loop CLI core

start spawn run

watch phase events

handoff human decision

resume same run

inspect evidence

Why it matters

Client choice should not own the workflow.

Teams experiment. A workflow might start in one agent client and later move to another. Without a boundary, status, approvals, evidence and recovery habits get trapped in whichever product the first adopter used.

Tools

start, watch, handoff, resume, inspect.

The core loop is small: start a run, watch meaningful changes, decide on handoff, resume the same run and inspect artifacts.

Resources

Evidence is reachable without log scraping.

MCP resources and tools can expose run summaries, findings, diffs, metrics and artifacts as structured surfaces.

Portability

Switch clients without rewriting the run model.

MCP is the practical boundary: clients can change, while lifecycle, handoff decisions, evidence and metrics remain Orcho-owned. TUI and web panel are in development / coming soon surfaces over that same model.