OpenClaw documents host-side ACP progress relay with `streamTo: "parent"`. MCACP exposes worker events, but same-chat rendering in Hermes is unverified. A Cursor-specific MCP bridge claims streaming; Hermes's progress-to-chat path still needs validation.

The worker should not know how to publish into the user's chat. The relevant feature is automatic host-side rendering of runtime events.

Closest documented ready implementation

OpenClaw supports external ACP harnesses and documents:

• sessions_spawn with runtime: "acp" and streamTo: "parent": initial-run progress summaries go back to the requester session.
• acp.stream.deliveryMode: "live": incremental delivery instead of waiting for completion.
• Visibility controls for status tags and parent progress. The sessions documentation says Discord parent progress needs explicit streaming.mode: "progress".

This is host-side behavior; it does not require a worker to call a chat tool. However, the documented parent relay is progress summaries, not a guarantee that every raw ACP event becomes an independently rendered chat item.

Sources: OpenClaw ACP session parameters, OpenClaw ACP runtime configuration.

This is evidence for a ready implementation in OpenClaw, not a verified Hermes integration or a recommendation to replace the orchestrator without further evaluation. Neither Hermes + OMP compatibility nor Cloudflare Durable worker support has been established here.

What the Hermes candidates actually establish

MCACP exposes intermediate worker events through asynchronous polling and batches message chunks. Its documentation does not establish automatic rendering into a Hermes conversation.

Source: MCACP configuration and interaction patterns.

A narrower project, tommulkins/cursor-hermes-bridge, says it converts ACP message chunks into MCP notifications/progress and claims Hermes displays partial output. It targets Cursor, auto-approves permissions, and lists concurrent sessions as unfinished. It is not a verified generic OMP orchestrator.

Source: cursor-hermes-bridge README.

An open Hermes PR reports that ordinary MCP calls do not request SDK progress notifications. Its proposed callback only refreshes the idle deadline; the author explicitly separates progress payload handling from that fix. This means a server emitting MCP progress is insufficient evidence that the payload reaches the user's chat. The Cursor bridge's display claim needs a version-specific end-to-end check.

Source: Hermes MCP progress timeout PR #94727.

The remaining Hermes gap is a verified runtime-to-conversation renderer. Worker-side hermes send and model-mediated polling/forwarding do not satisfy the clarified requirement.