AWS MCP Server
Connect AI agents to AWS — query and operate cloud resources across 15,000+ AWS API operations.
Updated
What is AWS MCP?
The AWS MCP server is a Model Context Protocol server, hosted and managed by Amazon Web Services, that gives AI agents and coding assistants secure access to query AWS documentation and operate live cloud resources across AWS services.
It is a managed remote server reached at https://aws-mcp.us-east-1.api.aws/mcp (with a Frankfurt endpoint, https://aws-mcp.eu-central-1.api.aws/mcp, also available). The endpoint itself requires no separate API key or signup — instead, requests are authenticated with your existing AWS IAM credentials. Because MCP clients speak OAuth while AWS uses SigV4, you connect through the open-source MCP Proxy for AWS, which signs each request with your IAM identity. This means the agent's permissions are exactly your IAM permissions, and you can constrain them further with IAM condition keys.
Beyond read-only documentation lookups, the server can execute real operations: the call_aws tool runs any of the 15,000+ AWS API operations, while run_script executes Python in a sandboxed environment for multi-step, cross-service workflows. It works with Claude Code, Cursor, Kiro, Claude Desktop, and any MCP-compatible client.
Tools the AWS MCP server exposes
| Tool | What it does |
|---|---|
| call_aws | Execute authenticated AWS API calls with syntax validation and error handling — supports most of the 15,000+ AWS APIs. |
| run_script | Run Python in a sandboxed environment with AWS API access for parallel calls, multi-step, and cross-service workflows. |
| get_presigned_url | Generate pre-signed Amazon S3 URLs for uploading or downloading files. |
| get_tasks | Poll the status of long-running tasks started by call_aws or run_script. |
| search_documentation | Search across all AWS documentation, including API references, best practices, service guides, and skills. |
| read_documentation | Retrieve and convert AWS documentation pages to markdown for easy consumption by AI assistants. |
| recommend | Get content recommendations for AWS documentation pages based on related and commonly viewed topics. |
| retrieve_skill | Retrieve domain-specific AWS expertise — workflows, best practices, decision frameworks, and step-by-step procedures. |
| list_regions | Retrieve a list of all AWS regions, including their identifiers and names. |
| get_regional_availability | Check AWS regional availability for services, features, SDK APIs, and CloudFormation resources. |
Connect the AWS MCP server
Claude Code
- 1
Install prerequisites
Install the AWS CLI (v2.32.0+) and uv. uvx runs the MCP Proxy for AWS that signs requests with SigV4.
- 2
Sign in to AWS
Run `aws login`, then verify with `aws sts get-caller-identity`. Your IAM credentials authorize every agent action.
- 3
Add the server
Run: claude mcp add aws-mcp -- uvx mcp-proxy-for-aws==1.6.2 https://aws-mcp.us-east-1.api.aws/mcp --metadata AWS_REGION=us-east-1
- 4
Verify
Restart Claude Code and run /mcp. Ask "What AWS regions are available?" to confirm the tools loaded.
Cursor
- 1
Install prerequisites
Install the AWS CLI (v2.32.0+) and uv, then run `aws login` to obtain IAM credentials.
- 2
Edit MCP config
In Cursor's mcp.json add an aws-mcp server with command "uvx" and args ["mcp-proxy-for-aws==1.6.2", "https://aws-mcp.us-east-1.api.aws/mcp", "--metadata", "AWS_REGION=us-east-1"].
- 3
Reload and verify
Reload Cursor, confirm aws-mcp shows tools like aws___call_aws and aws___search_documentation, then prompt the agent to query AWS.
Sample use cases
Security & permissions
The AWS MCP server endpoint requires no separate API key — but it is not unauthenticated. Every request is authenticated with your AWS IAM credentials via SigV4, bridged from MCP's OAuth through the open-source MCP Proxy for AWS. The agent therefore inherits exactly your IAM permissions: it can read and operate only the AWS resources your identity is allowed to. Scope this down with least-privilege IAM policies and IAM condition keys to restrict which actions an agent can take. aws login issues short-lived credentials that rotate automatically (default sessions up to 12 hours), limiting exposure. Note that call_aws and run_script can perform write/destructive operations, so grant only the permissions a given workflow needs.
Using AWS MCP with Gamut
In a Gamut workflow, an agent can use the AWS MCP server as a tool to inspect and act on cloud infrastructure on a schedule or in response to an event — for example, a scheduled run that audits S3 bucket settings, summarizes a CloudWatch metric, or checks resource availability across regions, then posts the result to Slack. Gamut manages the connection and AWS credentials so the agent calls tools like call_aws and search_documentation directly within the run. Because actions are bounded by the underlying IAM permissions, scope the connected identity to least privilege for the specific task.
Frequently asked questions
Which AWS services does the MCP server support?
The AWS MCP server typically covers core services like S3, EC2, Lambda, CloudWatch, IAM, and RDS. Coverage depends on the specific MCP server implementation you use.
Does it use my existing AWS credentials?
Yes. The MCP server authenticates with your AWS credentials (access key, IAM role, or SSO profile) — it inherits whatever permissions those credentials have.
Is there a risk of accidental infrastructure changes?
The MCP server executes the operations your agent requests. Use IAM policies to restrict write permissions, and consider read-only credentials for monitoring-only agents.
Can I scope access to specific AWS regions or accounts?
Yes. Configure the MCP server with region-specific endpoints or assume roles in specific accounts to limit the blast radius of agent operations.
Does it support multi-account AWS Organizations setups?
If your credentials support cross-account role assumption, the MCP server can operate across accounts. Configure trusted roles for each target account.
Are AWS API rate limits handled?
AWS services have per-service rate limits. The MCP server passes through API calls, so your agent should handle throttling (429/503 responses) with backoff logic.
Is the AWS MCP server open source?
There are several AWS MCP server implementations — some community-built and open source, and an official one from AWS. Check the specific repo for license and feature details.