Action Gateway Overview
Action Gateway gives agents catalog tools through a managed MCP endpoint using action_search, action_invoke, and action_code, with on-demand discovery, preloading, and output views to control model context.
Action Gateway Conceptspublic
Generated on 23 Sep 2026
Action Gateway gives applications and AI agents governed access to tens of thousands of tools through a managed MCP endpoint or SDK. See What You Can Build for example use cases of Action Gateway with Harness Runtime.
Sessions
Action Gateway sessions bind tool access, actor connections, policy, instructions, and optional private networking to an MCP endpoint.
Actors
An Action Gateway actor is a user-defined identity that links a user’s provider accounts to the sessions that use them.
Connections
Action Gateway connections authorize provider accounts for an actor, belong to the creating DigitalOcean user, and keep credentials out of model context.
Toolbelts
Action Gateway Toolbelts group tools into versioned collections for session selection, group permissions, and preloading.
Tool Policies
Action Gateway tool policies allow, require approval for, or deny calls within a session’s selected catalog.
Preloading Tools
Action Gateway preloading exposes known tools directly to the model while tool search discovers other relevant tools as needed.
Tool Search
Action Gateway tool search discovers relevant catalog tools and their schemas without loading the entire catalog into an agent’s context.
Invoke Tool
Action Gateway invokes named catalog tools with validated arguments and returns an outcome for each call.
Code Execution Tool
Action Gateway runs an agent’s Python code in an ephemeral sandbox, with optional programmatic tool calling for advanced workflows.
Reliable Execution
Action Gateway uses bounded, tool-specific retries and provider idempotency where supported, without guaranteeing that arbitrary repeated calls are safe.