AWS Documentation Model Context Protocol (MCP) Server

1.1.20 · active · verified Sun Apr 12

The AWS Documentation Model Context Protocol (MCP) Server is an open-source library from AWS Labs that provides real-time access to official AWS documentation for AI assistants and agents. It enables AI coding assistants to search, fetch, and get recommendations from up-to-date AWS documentation, reducing hallucinations and improving the accuracy of AI-generated content for cloud-native development. The MCP protocol itself is an open standard overseen by the Linux Foundation. The library is currently at version 1.1.20 and is actively maintained with a regular release cadence.

Warnings

breaking Server-Sent Events (SSE) support was removed from all MCP servers in their latest major versions as of May 26, 2025. Applications that still rely on SSE for client-server communication must use previous major versions of the respective MCP server until they can migrate to alternative transport methods.
Fix: Migrate client applications to use Streamable HTTP transport or other supported communication methods, or explicitly pin to an older major version of the server that still includes SSE support.
gotcha The server requires AWS credentials to function, typically picked up from the default AWS CLI profile or `AWS_PROFILE` environment variable. It also runs in read-only mode by default, restricting operations to only perform read actions. Write-access (for mutating operations like `sam_deploy`) requires explicit configuration (e.g., `enable_write_access_mode`).
Fix: Ensure AWS CLI is installed and configured (`aws configure` or `aws login`). Set `AWS_PROFILE` and `AWS_REGION` environment variables as needed. If write operations are required, consult the server's documentation for enabling write access mode.
gotcha The library explicitly requires Python 3.10 or newer. Attempting to run it with older Python versions will result in execution errors.
Fix: Upgrade your Python environment to version 3.10 or higher.
gotcha Running multiple AWS-related MCP servers (e.g., `aws-api-mcp-server`, `aws-knowledge-mcp-server`) concurrently in the same MCP client configuration can lead to tool conflicts, ambiguous responses for AI agents, and reduced performance. It's recommended to remove conflicting server entries.
Fix: Review your MCP client's configuration file (e.g., `~/.kiro/settings/mcp.json`, `~/.aws/amazonq/mcp.json`) and remove any entries for other AWS MCP servers that might overlap in functionality with the AWS Documentation MCP Server.

Install

pip install awslabs-aws-documentation-mcp-server Install with pip

Imports

Server Execution
```
python -m awslabs.aws_documentation_mcp_server.server
```
This library is designed to run as a server process, typically invoked as a module from the command line or configured by an MCP client, rather than being imported directly into user-written Python code for its core server functionality.

Quickstart

This quickstart demonstrates how to run the AWS Documentation MCP Server as a Python module. Before running, ensure you have Python 3.10+ installed and your AWS CLI credentials are configured (e.g., `aws configure`). The server will use these credentials to interact with AWS services for documentation retrieval. Once running, it can be integrated with compatible MCP clients like Amazon Q Developer, Kiro, or Cursor.

import os
import subprocess
import sys

# Configure AWS credentials (replace with your actual region/profile or ensure they are set in your environment)
os.environ['AWS_PROFILE'] = os.environ.get('AWS_PROFILE', 'default') # Use 'default' profile or env var
os.environ['AWS_REGION'] = os.environ.get('AWS_REGION', 'us-east-1') # Use 'us-east-1' or env var

print(f"Starting AWS Documentation MCP Server using profile: {os.environ['AWS_PROFILE']} in region: {os.environ['AWS_REGION']}")
print("Ensure your AWS credentials are configured (e.g., via `aws configure` or environment variables).")
print("The server will run until interrupted (Ctrl+C).")

try:
    # Run the MCP server as a module
    process = subprocess.run(
        [sys.executable, "-m", "awslabs.aws_documentation_mcp_server.server"],
        check=True
    )
except subprocess.CalledProcessError as e:
    print(f"Error starting server: {e}")
    sys.exit(1)
except KeyboardInterrupt:
    print("\nServer stopped by user.")
    sys.exit(0)

view raw JSON →