Documentation

Quickstart

Switchboard is an MCP gateway that connects your AI tools to 36+ integrations through a single endpoint. Get up and running in under 2 minutes — no infra, no config files, no CLI.

Hosted Setup

1. Sign up and create your org at app.switchboard-mcp.com
2. Add integrations from the dashboard (GitHub, Slack, Datadog, Google Workspace, and 30+ more)
3. Enter credentials or connect via OAuth — one click for most providers. Connecting all of Google Workspace takes a single sign-in.
4. Open the Install page in the dashboard for a copy-paste snippet tailored to your AI client

Authentication

Switchboard supports two ways to authenticate an MCP client, both scoped to your org and your policies:

• API key — create an sb_ key in the dashboard and send it as a bearer token. Best for CLIs and headless agents. Keys inherit your permissions and can be revoked at any time.
• OAuth 2.0 — for clients that support it, Switchboard is a full OAuth authorization server with Dynamic Client Registration and PKCE. The client walks you through sign-in in the browser; no long-lived key to copy or store.

Both paths enforce the same org membership, plan limits, and tool policies on every request. Use whichever your client supports — the endpoint URL is identical.

Connect Your AI Client

Cursor

{ "mcpServers": { "switchboard": { "url": "https://your-org.switchboard-mcp.com/mcp", "headers": { "Authorization": "Bearer sb_your_api_key" } } } }

Claude Desktop

{ "mcpServers": { "switchboard": { "url": "https://your-org.switchboard-mcp.com/mcp", "headers": { "Authorization": "Bearer sb_your_api_key" } } } }

Claude Code (CLI)

claude mcp add --transport http switchboard \ https://your-org.switchboard-mcp.com/mcp \ --header "Authorization: Bearer sb_your_api_key"

Codex, Windsurf, and VS Code use the same url + bearer-token pattern — grab a ready-made snippet for your client from the Install page in the dashboard.

Working with the Meta-tools

Switchboard doesn't expose 36 integrations as hundreds of raw tools. It exposes two:

• search — find the tools you need with a short keyword query (synonym-aware). Only the matching tools are returned, keeping your context window small.
• execute — run a tool by name, or chain several across integrations in one server-side script so intermediate results never round-trip through the model.

Set shared context once (like an owner and repo) and it's auto-injected into later calls; every result is pinned and can be referenced by handle. Most agents discover this workflow on their own — search first, then execute.