Skip to main content

What is Switchboard?

You've seen Connect give users a browser UI for editing documents. Switchboard is the other side of that coin — it runs the same engine, but on a Node server, and instead of a UI it hands clients an API. Same documents, same operation log, different surface.

Just another host​

Remember the host-application pattern: any runtime that boots a Reactor and loads a package is a host. Connect is one. Switchboard is another.

The difference is what each host exposes outward:

  • Connect renders editors — it gives people a UI.
  • Switchboard exposes GraphQL and MCP endpoints — it gives code an API.

Because both run the same Reactor library with the same document models installed, they stay in sync. An operation written through Connect arrives in Switchboard; a mutation dispatched via GraphQL shows up immediately in Connect. One Reactor, two surfaces.

What Switchboard exposes​

Switchboard serves two protocols at once:

  • GraphQL — composed from subgraphs, one per installed package. Web apps, mobile clients, and integrations query documents and dispatch actions here.
  • MCP (Model Context Protocol) — the same Reactor, surfaced as tools an AI agent can call.

You'll dig into subgraphs in the next lesson. For now, just think of them as "the GraphQL layer, split by package domain, then stitched back into one unified schema."

Where processors fit in​

Processors are event-driven handlers that subscribe to the operation stream and fold it into read models — optimised views for search indexes, dashboards, aggregates, or anything else that's awkward to derive from a single document's state at query time. Processors run inside Switchboard, and subgraphs query the read models they produce.

You'll get a full lesson on processors too. The short version: if a query spans many documents or needs pre-computed data, a processor builds it; a subgraph exposes it.

QuizA team is building three features. Match each one to the right tool.

Running Switchboard locally​

During development, ph vetra spins up a local Switchboard alongside Connect. It hot-reloads your subgraphs as you edit them and attaches MCP so you can test agent workflows without a production deployment.

ph vetra
# Reactor booted
# Subgraphs composed — GraphQL at http://localhost:4001/graphql
# MCP at http://localhost:4001/mcp

In production, Switchboard deploys as a Docker container — same code, persistent Reactor store, as many instances as you need behind a load balancer.