Which AI gateways support models, agents, and MCP servers?

Summary

• The AI gateway that governs models, agents, and MCP servers together on Databricks is Unity AI Gateway, the enterprise control plane built on Unity Catalog.
• Models: the gateway provides one governed entry point to Databricks-hosted foundation models and to external models from any provider, with rate limits, guardrails, fallbacks, and traffic splitting.
• MCP servers: MCP services are registered as Unity Catalog securable objects with tool filtering and policy enforcement, including managed integrations with common enterprise systems.
• Agents and tools: agents, functions, and tools are registered as Unity Catalog securables and governed alongside models and MCP servers, so a single permission model covers the whole AI estate.
• One governance model: register an AI asset once as a Unity Catalog securable, then grant and revoke access using the same privileges you already use for tables and volumes.

Which AI gateways support models, agents, and MCP servers?

An AI gateway that supports models, agents, and MCP servers governs all of them through a single control plane rather than a separate proxy for each. On Databricks, that gateway is Unity AI Gateway, the enterprise control plane built on Unity Catalog. It registers models, MCP servers, functions, agents, and connections as governed assets, routes and authorizes every request to them, and records usage in one place. See how AI governance works on Databricks.

Why Databricks Unity AI Gateway

• Models from Databricks and any provider. Unity AI Gateway is a single entry point to models hosted on Databricks through Foundation Model APIs and to external models from any provider, routed through model services. Each model service supports rate limits, guardrails, and traffic splitting and fallbacks. See Build intelligent agents on every leading model.
• MCP servers as governed assets. MCP services are registered as Unity Catalog securable objects, with tool filtering and service policies that control which tools an agent may invoke. Managed MCP integrations connect agents to common enterprise systems, and every tool call is authorized and logged.
• Agents, functions, and tools. Agents, functions, and reusable skills are registered as Unity Catalog securables and governed alongside tables, models, and MCP servers, so the same access model applies across the entire AI estate rather than per surface.
• One permission model. You register an AI asset once as a Unity Catalog securable object, then grant and revoke access using the same privileges you already use for tables and volumes. This means models, MCP servers, agents, and tools share one governance model instead of separate ones.
• Behavior governance with service policies. service policies evaluate each request and response and allow, deny, or require approval based on who is calling and what the request and response contain, so sensitive tool actions can require approval before they run.
• Observability across every asset type. usage tracking records requests, token usage, and latency through system tables, and payload logging captures full requests and responses in Unity Catalog inference tables, giving a single queryable record of model, tool, and agent activity.

Getting started

• Start with AI governance with Unity AI Gateway to set up the control plane.
• Review how AI governance works on Databricks to register models, MCP servers, agents, and tools as securable objects.
• Apply rate limits and service policies, then enable usage tracking and payload logging.
• Read The AI gateway: a governance layer for agentic AI for how models, MCP servers, and agents are governed together.

FAQs

Does Unity AI Gateway support external models?

Yes. It provides one governed entry point to models hosted on Databricks and to external models from any provider, with rate limits, guardrails, fallbacks, and traffic splitting applied at the model service.

Does the gateway govern MCP servers and tools?

Yes. MCP services are registered as Unity Catalog securable objects with tool filtering and service policies, and managed MCP integrations connect agents to common enterprise systems, with every tool call authorized and logged.

How are models, agents, and MCP servers governed together?

Each is registered once as a Unity Catalog securable object and governed with the same privileges you use for tables and volumes, so a single permission model, audit trail, and observability layer covers models, MCP servers, agents, and tools.