With the AWS MCP servers now generally available, DynamoDB has an official server maintained by AWS Labs — and it's more than a query pipe. It helps an agent both *design* tables (modelling access patterns, keys and indexes) and *operate* on data, all from natural language. If your stack runs on DynamoDB, here's how to wire it into Claude safely.
The pick: the AWS Labs DynamoDB server
Use the official awslabs.dynamodb-mcp-server, published to PyPI and part of the AWS Labs MCP collection. It's the one to build on — vendor-maintained, tracking the API, and shipped alongside servers for Aurora, DocumentDB, Keyspaces and the rest of the AWS database family. Run it locally over stdio; there's no service to host.
Step 1 — install with uvx
The cleanest install is uvx, which fetches and runs the server without a global pip install. In your Claude Desktop config:
{
"mcpServers": {
"dynamodb": {
"command": "uvx",
"args": ["awslabs.dynamodb-mcp-server@latest"],
"env": { "AWS_PROFILE": "your-profile", "AWS_REGION": "us-east-1" }
}
}
}
Step 2 — scope AWS credentials
This is the step that matters. The server acts with whatever IAM permissions your profile carries, so don't hand it an admin role. Create a dedicated IAM principal scoped to the specific tables and actions an agent should touch — and start read-only if you can. An agent that can DeleteItem across production tables is a bad day waiting to happen; least privilege is the whole game here.
Step 3 — verify
Restart Claude and ask: "List my DynamoDB tables and describe the key schema of each." Real tables back confirms the connection and your credentials resolve. From there you can ask it to model a new table for an access pattern, or run scoped reads against existing data.
DynamoDB vs the other database servers
DynamoDB's server leans into data modelling in a way the relational servers don't — because getting keys and indexes right up front is where NoSQL projects live or die. If your data is relational, compare the best MCP server for a database and the MongoDB server for the document-store equivalent.
Going further
For the broader AWS surface, see the AWS MCP server setup. Keep credentials and blast radius under control with MCP permission scoping patterns and MCP security best practices. Building a data stack? Browse the databases category or the data-analyst loadout.