AWS publishes an official suite of MCP servers (the awslabs/mcp project) — for AWS documentation, the AWS API, pricing, infrastructure-as-code, specific services and more. Some are fully managed remote servers; others run locally with uvx. Here's how to connect them.
Option A — managed remote servers (zero setup)
AWS offers managed remote servers like the AWS Knowledge MCP (and an AWS API server in preview) — no infrastructure to run. Add them as a remote/OAuth connector in a client that supports remote MCP. Best for documentation and read-only knowledge.
Option B — local servers via uvx
Most awslabs servers run locally. The docs server, for example:
{
"mcpServers": {
"aws-docs": {
"command": "uvx",
"args": ["awslabs.aws-documentation-mcp-server@latest"]
}
}
}
For servers that hit your account (API, cost, IaC), they use your standard AWS credentials chain (env vars or ~/.aws/credentials), so configure the AWS CLI first. Requires Python 3.10+ with uv. Restart the client; see MCP config file location.
Pick the right server
awslabs/mcp is a suite, not one server — install only what you need: documentation, API, pricing, CDK/Terraform (IaC), or service-specific ones. Adding all of them blows past Cursor's ~40-tool cap.
Verify
Ask: "What's the latest guidance for S3 bucket encryption?" (docs server) or "List my Lambda functions in us-east-1." (API server, needs credentials).
Security — important
Servers that touch your account act with your AWS permissions. Use a least-privilege, read-only IAM role/profile for the MCP, never your admin keys, and be deliberate about any write-capable server. An AI with broad AWS write access is a serious risk. See MCP security best practices and how to vet an MCP server.
Going further
AWS servers pair with GitHub and Sentry for a cloud-ops loop. Browse the devops category or curated loadouts.