On May 6, 2026, AWS announced the general availability of its MCP server. This is one of the largest cloud providers officially shipping a managed MCP integration, and it changes what AI coding agents can do with AWS infrastructure.

Here is what shipped, what it means for developers building with MCP, and where this fits in the broader ecosystem.

What the AWS MCP Server Does

The AWS MCP server gives AI agents secure, auditable access to AWS services through the Model Context Protocol. One server, any AWS API.

Unified API access. Agents call any AWS API through a single MCP tool. That includes operations requiring file uploads and long-running execution. No need to wire up individual service integrations.

Sandboxed Python execution. Agents can run Python code against AWS services for multi-step operations. The sandbox has no access to your local filesystem or shell tools. This is how agents handle complex workflows that require chaining multiple API calls with logic in between.

Agent Skills. AWS introduced a concept called Skills: curated guidance packages that agents discover and load on demand. Think of them as structured SOPs that keep context window usage low. Instead of stuffing a prompt with AWS documentation, the agent pulls only what it needs for the current task.

Credential-free discovery. Documentation search and Skill discovery work without AWS credentials. Agents can explore what is available before authenticating. Authentication is only required when the agent takes action on your infrastructure.

IAM guardrails. Every agent action runs through standard IAM policies. CloudWatch metrics and CloudTrail logging give you full visibility into what agents did, when, and with what permissions.

Pricing

No additional charge for the MCP server itself. You pay for the AWS resources your agents use, the same as if a human made those API calls. Available in US East (N. Virginia) and Europe (Frankfurt) at launch.

Why This Matters for the MCP Ecosystem

AWS is the third major cloud provider to ship official MCP support in 2026. Google Cloud announced MCP as a first-class primitive at Cloud Next in early May. Red Hat launched an MCP gateway in technology preview the same week, built for traffic control between agents and MCP servers at the infrastructure layer.

The pattern is clear: MCP is moving from developer tooling to cloud infrastructure. When AWS, Google, and Red Hat all ship MCP integrations in the same month, the protocol is no longer experimental. It is becoming the expected interface between agents and cloud services.

For developers, this means a few things:

  1. Fewer custom integrations. If your agent needs to interact with AWS, you point it at the official MCP server. No custom API wrappers. No prompt engineering around service-specific quirks.

  2. Governance comes built in. IAM policies, CloudTrail audit logs, and sandboxed execution are there from day one. That matters when your agent is creating EC2 instances or modifying S3 buckets on your behalf.

  3. The directory is growing fast. Between AWS, Google Cloud, and the 250+ servers already indexed on AgentNDX, the MCP ecosystem crossed a density threshold where most common developer workflows have a server available.

How to Get Started

Install the AWS MCP server in any MCP-compatible client (Claude Desktop, VS Code with Copilot, Cursor, or your own agent):

npx @anthropic-ai/create-mcp --preset aws-mcp

Or configure it manually following the AWS Agent Toolkit documentation.

You will need AWS credentials configured locally (the standard ~/.aws/credentials or environment variables). IAM policies control what the agent can and cannot do.

FAQ

Q: Does this replace the AWS CLI for agents? A: It does not replace the CLI, but it is purpose-built for agent workflows. The MCP server handles context management, sandboxed execution, and credential scoping in ways the CLI was not designed for.

Q: Can I restrict which AWS services an agent can access? A: Yes. Standard IAM policies apply. You can scope an agent to read-only S3 access, for example, and the MCP server enforces that at the API level.

Q: Is this only for coding agents? A: The initial launch targets coding agents, but the architecture supports any agent that needs to interact with AWS services. DevOps automation, infrastructure provisioning, and monitoring workflows are all viable use cases.