AWS API MCP Server

The AWS API MCP Server is an official AWS Labs project that enables AI assistants to interact with AWS services through AWS CLI commands. It acts as a bridge between an MCP client (Claude Desktop, Cursor, VS Code, etc.) and the full AWS API surface, with command validation, security guardrails, and support for AWS services and parameters released after the model's knowledge cutoff.

The server exposes three tools: call_aws for executing validated CLI commands, suggest_aws_commands for translating natural language into CLI suggestions, and an experimental get_execution_plan for multi-step task guidance via agent scripts. It supports read-only mode, mutation consent prompts, custom security policies, and a default working-directory sandbox for file operations.

Because it is maintained by AWS Labs and shipped as part of the official awslabs/mcp monorepo (alongside many other AWS MCP servers), it offers broader and more current AWS coverage than community-built alternatives. It is intended for single-user, local use via stdio; an HTTP transport mode is available but adds significant attack surface.

Tools

Prerequisites

• AWS account with configured credentials (env vars, AWS profile, or IAM role)
• Python 3.10 or newer
• Optional: uv package manager for faster install

Install with uv (recommended, Linux/macOS)

Add to your MCP client config (e.g. Claude Desktop claude_desktop_config.json):

{
  "mcpServers": {
    "awslabs.aws-api-mcp-server": {
      "command": "uvx",
      "args": ["awslabs.aws-api-mcp-server@latest"],
      "env": {"AWS_REGION": "us-east-1"},
      "disabled": false,
      "autoApprove": []
    }
  }
}

Install with pip

pip install awslabs.aws-api-mcp-server

{
  "mcpServers": {
    "awslabs.aws-api-mcp-server": {
      "command": "python",
      "args": ["-m", "awslabs.aws_api_mcp_server.server"],
      "env": {"AWS_REGION": "us-east-1"},
      "disabled": false,
      "autoApprove": []
    }
  }
}

Install with Docker

{
  "mcpServers": {
    "awslabs.aws-api-mcp-server": {
      "command": "docker",
      "args": [
        "run", "--rm", "--interactive",
        "--env", "AWS_REGION=us-east-1",
        "--volume", "/full/path/to/.aws:/app/.aws",
        "public.ecr.aws/awslabs-mcp/awslabs/aws-api-mcp-server:latest"
      ]
    }
  }
}

Key environment variables

• AWS_REGION (default us-east-1): default region for CLI commands
• AWS_API_MCP_PROFILE_NAME: AWS named profile to use
• READ_OPERATIONS_ONLY (default false): block all mutating commands
• REQUIRE_MUTATION_CONSENT (default false): require explicit consent for writes
• AWS_API_MCP_WORKING_DIR: server working directory (sandbox for file ops)
• AWS_API_MCP_ALLOW_UNRESTRICTED_LOCAL_FILE_ACCESS: workdir | unrestricted | no-access
• EXPERIMENTAL_AGENT_SCRIPTS: enable get_execution_plan tool
• AWS_API_MCP_TRANSPORT: stdio (default) or streamable-http

Authentication uses the standard boto3 credential chain (env vars, ~/.aws/credentials, IAM role). For HTTP mode, set AUTH_TYPE=no-auth or AUTH_TYPE=oauth with AUTH_ISSUER and AUTH_JWKS_URI.

• Inspect AWS account inventory (EC2 instances, S3 buckets, Lambda functions, RDS clusters) on demand from a chat interface
• Diagnose production issues by tailing CloudWatch logs, checking ECS task states, and describing failed resources via natural language
• Provision and modify infrastructure (create IAM roles, launch EC2 instances, update Lambda config) with mutation-consent guardrails
• Translate plain-English requests like "show me unencrypted S3 buckets" into the correct AWS CLI command without manual lookup
• Run read-only audits and compliance checks safely by setting READ_OPERATIONS_ONLY=true

• "List all EC2 instances in us-west-2 that have been running for more than 30 days"
• "Show me the last 50 error log events from the /aws/lambda/my-function CloudWatch log group"
• "Suggest the AWS CLI command to enable versioning on the S3 bucket my-app-uploads"
• "Find any IAM users without MFA enabled in this account"
• "Create a new SQS queue named order-events with a 5 minute visibility timeout"

• Official AWS Labs project, kept current with new AWS services and parameters
• Universal coverage: any AWS CLI command works, not just a hand-picked subset
• Built-in safety controls: read-only mode, mutation consent, custom security policies, working-directory sandbox
• Multiple install paths (uvx, pip, Docker) and supports both stdio and streamable-HTTP transports

• Requires Python 3.10+ and local AWS credentials; not a managed remote server
• Explicitly not designed for multi-tenant or shared deployments, server runs with the launching user's permissions and no sandbox
• Pattern matching and wildcards are not supported in security policies, exact command names must be specified
• get_execution_plan is experimental and subject to change

• AWS CDK MCP Server (also from awslabs/mcp) for infrastructure-as-code workflows
• AWS CloudFormation MCP Server (awslabs/mcp) for template-based provisioning
• Community AWS MCP implementations on GitHub, though coverage and maintenance vary