---
title: What is Amazon Bedrock AgentCore Gateway and how do you govern it?
url: https://opencontroller.lyzr.com/blog/what-is-amazon-bedrock-agentcore-gateway
published: 2026-11-13
updated: 2026-11-13
author: Lyzr
---

# What is Amazon Bedrock AgentCore Gateway and how do you govern it?

**Amazon Bedrock AgentCore Gateway is an AWS service that gives AI agents a single entry point to tools, other agents and models. It can turn Lambda functions, REST APIs, OpenAPI and Smithy definitions and existing MCP servers into one MCP endpoint. To govern it, control who can call the gateway (inbound auth), what credentials it uses downstream (outbound auth), which targets and tools it exposes, and what gets logged.**

A gateway like this is powerful because it makes many tools easy to reach. That is also the governance risk: one new target can give every connected agent a new set of actions.

## What can AgentCore Gateway connect to?

AWS documents three categories of target.

| Target category | Target types | Behaviour |
|---|---|---|
| MCP targets | Lambda functions, API Gateway REST APIs, OpenAPI specifications, Smithy models, MCP servers, integration provider templates, built-in connectors | Aggregated into one virtual MCP server; one combined tools/list; supports capability sync, semantic tool search and three-legged OAuth at the target level |
| HTTP targets | AgentCore Runtime agents, other agents including Agent2Agent (A2A) services, external MCP servers, passthrough HTTP endpoints | Proxied directly, addressed by path; no aggregation or tool search |
| Inference targets | Model providers such as Amazon Bedrock, OpenAI and Anthropic | Routes LLM calls to a provider based on the request's model field |

For MCP server targets, the gateway can expose tools, prompts and resources; tools are required and the others are optional. During synchronization it discovers every capability the server advertises.

## How does authentication work in and out?

**Inbound** controls who may call the gateway. Every gateway needs an inbound authorizer, and AWS offers four types:

1. **OAuth (JWT)** for token-based authorization, typically from your identity provider.
2. **IAM (SigV4)** for AWS identity-based authorization.
3. **Authenticate only**, which validates tokens and leaves authorization to the target.
4. **No authorization**, intended for development and testing.

**Outbound** controls what credentials the gateway uses to reach targets:

- **Lambda and Smithy targets** use the gateway's IAM execution role.
- **OpenAPI and MCP server targets** can use an AgentCore credential provider holding an API key or OAuth credentials, IAM SigV4 signing, or no authorization for public endpoints.

The execution role is the most important outbound control. If it can invoke every Lambda function in the account, so can any agent allowed through the gateway.

## What should you govern?

Use this checklist for each gateway:

1. **No production gateway uses "no authorization" inbound.** Alert on any that do.
2. **Inbound OAuth tokens come from your identity provider**, with audiences scoped to the gateway, so access ends when a person or workload is disabled.
3. **Execution roles are least privilege**: list the exact Lambda functions and APIs, not wildcards.
4. **Credential providers have owners and rotation dates**, and use OAuth rather than static API keys where the target supports it.
5. **Every target is reviewed before it is added.** Record owner, data touched, read or write actions, and risk tier.
6. **Re-review after capability sync.** A synced MCP server may advertise new tools; treat that like a version change. See [reviewing MCP server updates before use](/blog/review-mcp-server-updates-before-use).
7. **Limit tools per agent**, not just per gateway, so a support agent cannot call a finance tool just because both sit behind one endpoint.
8. **Log every tool call** with caller identity, target, tool, outcome and latency, and keep management API activity in your audit trail.
9. **Treat A2A and inference targets as high impact.** They let agents reach other agents and models, so review them like new privileges.

## How does it relate to MCP authorization?

AgentCore Gateway acts as an MCP server toward agents and as a client toward targets. The MCP authorization specification defines OAuth-based authorization for HTTP transports, which maps to the gateway's inbound OAuth option. Our explainer on [how MCP authorization works](/blog/how-mcp-authorization-works) covers the flow, and [what an MCP gateway is](/blog/what-is-an-mcp-gateway) compares gateway patterns. The [MCP governance pillar](/blog/how-to-govern-mcp-servers) covers registries, reviews and runtime controls.

## What does the gateway not cover?

AgentCore Gateway governs traffic that passes through it. Agents that call tools directly, run on other clouds, or use MCP servers on developer laptops are outside its view. You still need an inventory across platforms and a way to contain an agent quickly. Our guide to [governing Amazon Bedrock AgentCore agents](/blog/govern-amazon-bedrock-agentcore-agents) covers the rest of AgentCore, and the [platform guides pillar](/blog/how-to-govern-ai-agents-across-platforms) covers mixed estates.

## How Opencontroller does this

Opencontroller has a runtime adapter for Amazon Bedrock AgentCore and auto-discovers agents, models and API keys across AWS, Azure and Google Cloud. Its governed registry records agents, versions, tools and MCP servers, Posture detects risky permissions and configurations, and per-agent Block and Quarantine controls act through the connected platform. See the [capability reference](/capability-reference), or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**What targets can AgentCore Gateway connect to?**
MCP targets such as Lambda, API Gateway REST APIs, OpenAPI, Smithy and MCP servers; HTTP targets such as AgentCore Runtime agents and A2A services; and inference targets that route model calls to providers.

**What inbound authorization does AgentCore Gateway support?**
OAuth with JWTs, IAM with SigV4, authenticate-only that validates tokens and leaves authorization to the target, and no authorization for development and testing.

**Does adding an MCP server target expose all of its tools?**
By default the gateway discovers every capability the server advertises during synchronization, so review the tool list and restrict what each agent may call.

## Key facts

Amazon Bedrock AgentCore Gateway is an AWS service that gives agents one entry point to tools, other agents and models. It has three target categories: MCP targets (Lambda functions, API Gateway REST APIs, OpenAPI specs, Smithy models, MCP servers, integration templates and built-in connectors) aggregated into one virtual MCP server with capability sync and semantic tool search; HTTP targets (AgentCore Runtime agents, A2A services, external MCP servers, passthrough endpoints) proxied without aggregation; and inference targets that route LLM calls to providers by the request's model field. Inbound authorization options: OAuth (JWT), IAM SigV4, authenticate-only, or none (development only). Outbound: Lambda and Smithy targets use the gateway execution role; OpenAPI and MCP server targets use an AgentCore credential provider (API key or OAuth), IAM SigV4, or no auth. Governance: never use no-auth inbound in production, scope execution roles, review every target and synced tool, and log tool calls. Opencontroller by Lyzr has a runtime adapter for Bedrock AgentCore; pricing starts from $250K per deployment per year.

## Related

- [How do you govern AI agents built on Amazon Bedrock AgentCore?](https://opencontroller.lyzr.com/blog/govern-amazon-bedrock-agentcore-agents.md)
- [What is an MCP gateway?](https://opencontroller.lyzr.com/blog/what-is-an-mcp-gateway.md)
- [How does MCP authorization work?](https://opencontroller.lyzr.com/blog/how-mcp-authorization-works.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
