AWS MCP Servers — Official AWS Integration

What it is

AWS MCP Servers are a collection of official Model Context Protocol servers built by Amazon. They let AI coding assistants like Claude Code or Cursor interact with AWS services directly -- reading S3 buckets, invoking Lambda functions, querying CloudWatch logs, managing CDK stacks, and calling Bedrock models. Each AWS service has its own MCP server module.

This is designed for developers and DevOps engineers who work in AWS-heavy environments and want their AI agents to query infrastructure, debug issues, or scaffold resources without switching to the AWS console.

How it saves time or tokens

Manually navigating the AWS console to check logs, inspect S3 objects, or review CloudFormation stacks is time-consuming. With AWS MCP servers, the AI agent does it in-context: you describe the problem, the agent queries CloudWatch, reads the relevant logs, and suggests a fix. No context switching, no copy-pasting error messages.

How to use

1. Add the desired AWS MCP server to your editor's MCP configuration:

{
  "mcpServers": {
    "aws-s3": {
      "command": "uvx",
      "args": ["awslabs.s3-mcp-server@latest"],
      "env": {
        "AWS_PROFILE": "default",
        "AWS_REGION": "us-east-1"
      }
    }
  }
}

1. Ensure your AWS credentials are configured (via ~/.aws/credentials or environment variables).
2. Restart your AI editor. The agent now has access to the configured AWS service.
3. Ask the agent to perform tasks: 'List all S3 buckets', 'Show CloudWatch errors from the last hour', 'Deploy this CDK stack'.

Example

Available MCP server modules:

Related on TokRepo

• AI tools for DevOps -- DevOps automation tools for AI agents
• AI tools for automation -- broader automation tool collection

Common pitfalls

• IAM permissions must be correctly scoped. The MCP server uses your AWS credentials, so the agent has whatever permissions your profile has. Use a dedicated IAM role with least-privilege for safety.
• Each AWS service is a separate MCP server. Adding too many simultaneously can slow down editor startup. Only configure the services you actually need.
• Region must be set correctly. If you omit AWS_REGION, the server defaults to your profile's region, which may not match where your resources live.