JA EN
shiichan

The AWS MCP Server Hits GA: Hand Your AI Agents the Latest AWS

AWS Blog ★★★★ View Markdown
#aws#ai#devops

Hey there, it's Shii-chan! Today I've got news that AWS folks have been waiting for: working with AWS from an AI agent just got a whole lot smoother.

AWS Blog aws.amazon.com

What was announced?

On the AWS Blog, AWS announced that the AWS MCP Server is now generally available (GA)! It's a managed, remote Model Context Protocol server that gives AI agents and coding assistants authenticated access to AWS services. It's part of the Agent Toolkit for AWS, and it lets your agent operate AWS through a curated set of tools.

Why it matters

Until now, asking an AI agent to do AWS work could get bumpy. Training data goes stale so agents miss newer services, they lean on the CLI instead of infrastructure-as-code, and they tend to over-broaden IAM policies, producing setups that aren't production-ready. The AWS MCP Server is built to close that gap between agents and AWS.

What changes

Your agent can now reach the "current" state of AWS whenever it needs to. It can read documentation on the spot, so it's far less likely to answer from outdated knowledge, and it acts within your IAM permissions. In the blog's example, Claude Code first suggested an outdated S3 embedding approach, but once the AWS MCP Server was enabled it correctly pointed to Amazon S3 Vectors. That's the power of real-time documentation.

Dive Deep

The core tools look like this:

  • call_aws: run any of 15,000+ AWS API operations using your existing IAM credentials
  • search_documentation and read_documentation: fetch up-to-date AWS docs on the spot (in GA, this no longer requires authentication)
  • run_script: run Python scripts in a sandboxed environment that inherits your IAM permissions but has no network access

GA also adds support for IAM context keys, reduced token consumption, and a clear separation between human and agent permissions. On top of that, metrics land in a dedicated AWS-MCP CloudWatch namespace, and CloudTrail records every API call, so auditing is covered too.

Setup uses the MCP Proxy for AWS, which bridges IAM authentication to OAuth 2.1. With Claude Code you can add it like this:

claude mcp add-json aws-mcp --scope user \
'{"command":"uvx","args":["mcp-proxy-for-aws==1.6.0","https://aws-mcp.us-east-1.api.aws/mcp","--metadata","AWS_REGION=us-west-2"]}'

Client support is broad: Claude Code, Kiro, Cursor, Codex, and pretty much any MCP-compatible tool.

It's available in US East (N. Virginia) and Europe (Frankfurt), and it can make API calls to any region. There's no charge for the MCP Server itself; you only pay for the AWS resources you create and data transfer. For more, check the AWS MCP Server User Guide.

Wrap-up

  • The AWS MCP Server is GA: a managed, remote MCP server that gives AI agents authenticated access to AWS
  • A curated toolset (call_aws, documentation retrieval, run_script) tackles stale knowledge and overly broad permissions
  • Observability and auditing are covered via the AWS-MCP CloudWatch namespace and CloudTrail
  • The server itself is free, available in US East (N. Virginia) and Europe (Frankfurt)
  • This one lands hardest for developers who want to work with AWS from AI agents and coding assistants!

Related posts