Skip to main content
This guide walks through the three-layer integration model in AgentRuntime: connections (credentials), MCP instances (deployed tool servers), and workflow steps (runtime tool calls).
Docs structure: Pages in Integration setup cover the shared model (connections, instances, auth). Connectors lists one page per MCP adapter (connectors/{slug}) with tools and config only for that adapter.

Before you begin

  • Project contributor role in your workspace
  • Credentials for the tool you want to connect (Google account, API key, or WhatsApp business account)

The integration model

  1. Connection — Stores OAuth tokens or API keys
  2. MCP instance — Installs a catalog server in your project and binds a connection
  3. Workflow step — Calls a tool on that instance at runtime

Steps

1

Create a connection

Go to Connections (/connections) and connect your tool:
  • Google account card — OAuth for Workspace MCP adapters (Gmail, Drive, Sheets, …)
  • Connect WhatsApp — Embedded signup or manual token entry
  • New custom connection — API keys, tokens, DSNs (Resend, GitHub, Postgres, …)
See Connections.
2

Install an MCP instance

Go to MCP (/mcp), find a server in the platform catalog (for example, gmail, postgres, or shopify), and click Add instance.
3

Wire credentials on the instance

Open Instance config (/mcp/instances/{id}). Pick a config profile, wire your connection, and set the profile active.
4

Validate (recommended)

Run MCP validation at /mcp/{server_id}/validate or test with a single-tool workflow run before production traffic.
5

Add an MCP step to a workflow

In Workflow Studio, add an mcp_call step. Select your instance, pick a tool, and configure tool_args. Reference upstream data with template variables.
6

Validate and run

Dry-run validate the workflow, publish, and start a run. Check the event log for tool responses.

Next steps