MCP editor for agents
MCP editor for agents
Use a dedicated MCP transport so AI agents can read published vs workspace drafts and edit tools safely without publishing.
Overview
Rapid MCP exposes an editor MCP endpoint alongside your normal server stream. Agents connect with the same authentication as production, but get a small, purpose-built tool surface to:
- Compare the published snapshot with the workspace (draft) snapshot
- Create or update tools on the workspace draft only (nothing is published until you publish in the dashboard)
When an agent changes a tool, the dashboard can refresh automatically: the backend notifies connected browsers over the same Socket.IO channel used elsewhere, so the manage-server UI stays in sync with draft edits made from Cursor or other MCP clients.
Video walkthrough
Why it exists
- Published vs draft: Production traffic should keep using immutable published versions. Agents need to see what’s live and what you’re still editing.
- Scoped writes: The editor transport is limited to draft tool upserts (no surprise publishes).
- Operator UX: Editors working in the web UI see updates when an agent changes the draft, without manual refresh.
For background on publishable versions, see Environments & versions.
Endpoint
Use the editor stream URL configured in your MCP client (same host and auth as your main MCP server, with the editor path your deployment exposes—typically the stream/edit route on the MCP backend).
Configure it like any other MCP server entry (URL + Authorization headers, etc.). Your JWT and server id must allow access, consistent with Authentication.
Dashboard sync
After a successful draft tool upsert from the editor MCP, Rapid MCP emits a SERVER_CHANGED-style event to the server room so open manage server tabs refetch and show the new draft state.
Related
- Environments & versions — published snapshots vs workspace
- MCP tools — tool schema and behavior
- Connect to agents — URL, headers, and clients