Set up the AWS MCP Server

The AWS MCP Server is a managed Model Context Protocol server from the Agent Toolkit for AWS. It gives your AI agent authenticated access to AWS – running AWS API calls on your behalf, executing scripts in a sandboxed environment, and serving AWS skills (including the amazon-opensearch-service skill) on demand. Use it to work with OpenSearch Service domains and OpenSearch Serverless collections from your agent.

Prerequisites

Install the aws-data-analytics plugin (recommended)

The aws-data-analytics plugin bundles the AWS MCP Server configuration and the amazon-opensearch-service skill in a single install, so you do not have to configure the MCP endpoint or install the skill separately.

Claude Code

/plugin install aws-data-analytics@claude-plugins-official
/reload-plugins

Codex

codex plugin marketplace add aws/agent-toolkit-for-aws

Then launch Codex and run /plugins to browse and install the aws-data-analytics plugin.

Configure the AWS MCP Server directly

If your agent does not support plugins, configure the AWS MCP Server directly. Follow Setting up the AWS MCP Server in the Agent Toolkit for AWS documentation, then ask your agent an OpenSearch question – it discovers the amazon-opensearch-service skill at runtime through the server.

Use the AWS MCP Server without skills

You can use the AWS MCP Server directly without loading the amazon-opensearch-service skill. In this mode, the agent uses the server's built-in call_aws tool to invoke AWS API operations directly. This is useful for ad-hoc operations or when you want precise control over the API calls your agent makes.

For example, to create an OpenSearch Serverless collection:

Create an OpenSearch Serverless search collection named "my-collection" in us-east-1

The agent calls the OpenSearch Serverless CreateCollection API on your behalf using your configured credentials:

{
  "name": "my-collection",
  "type": "SEARCH",
  "description": "My search collection"
}

You can also run operations against existing resources without skills. For example:

List all my OpenSearch Service domains in us-east-1
Show the cluster health for my domain named production-domain
Create an index called products in my collection endpoint

The AWS MCP Server handles authentication and translates your natural language requests into the appropriate AWS API or OpenSearch API calls.

Verify the setup

After installing, restart your agent so it loads the new configuration. Then ask:

What AWS skills do you have available for OpenSearch?

The agent should report the amazon-opensearch-service skill. Try a task such as "List my OpenSearch Service domains" – the agent runs the call through the AWS MCP Server using your configured credentials.

Security considerations

The AWS MCP Server runs with the credentials you provide. Follow these practices:

Troubleshooting

Use with agent frameworks

You can integrate the AWS MCP Server into Python agent frameworks for programmatic workflows. The following examples connect to the AWS MCP Server endpoint, which provides authenticated access to OpenSearch and other AWS services.

Strands Agents

Strands Agents is an AWS-native agent SDK with built-in MCP support and Amazon Bedrock as the default model provider.

from strands import Agent
from strands.tools.mcp import MCPClient
from mcp_proxy_for_aws.client import aws_iam_streamablehttp_client

ENDPOINT = "https://aws-mcp.us-east-1.api.aws/mcp"
AWS_REGION = "us-east-1"

opensearch_client = MCPClient(
    lambda: aws_iam_streamablehttp_client(
        endpoint=ENDPOINT,
        aws_service="aws-mcp",
        aws_region=AWS_REGION,
    )
)

with opensearch_client:
    agent = Agent(tools=opensearch_client.list_tools_sync())
    response = agent("What OpenSearch Service domains do I have?")
    print(response)

Install the required packages:

pip install strands-agents mcp-proxy-for-aws

LangGraph

LangGraph is a low-level orchestration framework for building stateful agents. The following example uses mcp-proxy-for-aws to provide an authenticated transport and langchain-mcp-adapters to load the AWS MCP tools into a LangChain agent backed by Amazon Bedrock.

import asyncio
from mcp import ClientSession
from mcp_proxy_for_aws.client import aws_iam_streamablehttp_client
from langchain_mcp_adapters.tools import load_mcp_tools
from langchain_aws import ChatBedrock
from langchain.agents import create_agent

ENDPOINT = "https://aws-mcp.us-east-1.api.aws/mcp"
AWS_REGION = "us-east-1"

async def main():
    async with aws_iam_streamablehttp_client(
        endpoint=ENDPOINT,
        aws_service="aws-mcp",
        aws_region=AWS_REGION,
    ) as (read_stream, write_stream, _get_session_id):
        async with ClientSession(read_stream, write_stream) as session:
            await session.initialize()
            tools = await load_mcp_tools(session)
            model = ChatBedrock(
                model_id="us.anthropic.claude-opus-4-8",
                region_name=AWS_REGION,
            )
            agent = create_agent(model, tools)
            result = await agent.ainvoke(
                {"messages": [{"role": "user", "content": "What OpenSearch Service domains do I have?"}]}
            )
            print(result["messages"][-1].content)

asyncio.run(main())

Install the required packages:

pip install mcp-proxy-for-aws langchain langchain-aws langchain-mcp-adapters langgraph

More information