What Is AWS Bedrock MCP and What Does It Do?

The aws bedrock MCP server lets an agent query Amazon Bedrock Knowledge Bases in plain language. It is a retrieval tool, not a model gateway: the agent asks a question and gets answers drawn from your own enterprise data. The CheckMCP audit scores the repository 100/100, grade A. Installing takes one command, and the permissions you grant matter more than the install.

What is the aws bedrock MCP server?

The server listed as AWS Bedrock KB is a Cloud Platforms MCP server that lets Claude Code, Cursor, Windsurf and any MCP-compatible agent query Amazon Bedrock Knowledge Bases using natural language. Its own description sums it up as RAG over your enterprise data. In other words, the documents you already indexed in a knowledge base become something an agent can ask questions about.

The name causes some confusion, so it helps to be precise. The aws bedrock server here is the knowledge-base retrieval server, published in the awslabs/mcp repository and run through the package awslabs.bedrock-kb-retrieval-mcp-server. It is not a general wrapper around every Bedrock feature, and it is not the same thing as the broader AWS server, which covers storage, compute and logging.

As of November 5, 2026, the CheckMCP audit of awslabs/mcp reports a score of 100/100 (grade A), 9,874 stars, an Apache-2.0 licence, 253 open issues and a last commit dated the same day. The audit scores licence, adoption, maintenance and documentation, and it shows the grade whether or not it flatters the server. A high score says the repository is well kept. It does not say the server is right for your data.

What can an agent do with it?

The core capability is a single idea: the agent sends a natural-language question, and the server queries an Amazon Bedrock Knowledge Base on its behalf. That puts company material, such as internal documentation, policies or product notes, within reach of an agent without pasting it into the prompt. The agent can fetch what it needs when it needs it.

In practice this suits a few jobs:

• Answering questions inside Claude Code or Cursor from internal documents that were already indexed in a knowledge base.
• Grounding a coding or support task in your own reference material rather than the model's general knowledge.
• Checking what the knowledge base actually holds before an agent drafts something that relies on it.

These are uses that follow from the server's stated purpose, querying knowledge bases with natural language. The listing does not publish a tool-by-tool schema for this server, so confirm the exact tool names and inputs in the repository before you build a workflow around them. If you want to compare retrieval options first, Best MCP Servers for Vector Search and RAG in 2026 and What Is Pinecone MCP and What Can an Agent Do? cover other routes.

How do you connect it?

The recommended route is the MCPizy command line. Run mcpizy install aws-bedrock and the client configuration is written for you. The listing says installation takes about 1 minute and is marked as verified. If you prefer to wire it up yourself, the alternative the listing gives is to run the server directly with uvx awslabs.bedrock-kb-retrieval-mcp-server, which needs uv installed.

The listing shows the server working with Claude Code, Claude Desktop, Cursor, Windsurf, VS Code with Copilot and any other MCP client. The MCP configuration is the same idea everywhere, and only the config file path differs. For the VS Code side, How to Add an MCP Server to VS Code walks through the file, and VS Code mcp.json: the full schema, with examples covers its structure.

Before the first call, the server needs AWS credentials and a knowledge base to talk to. Those details, including which environment variables and region settings it reads, live in the repository's README, so follow it rather than copying a configuration from a different AWS server. Then ask the agent a question that your knowledge base should be able to answer, and check that the reply reflects your documents.

What does it not do?

It is not a model gateway. The aws bedrock server described here retrieves from knowledge bases. It does not, according to its own description, route prompts to models hosted on Bedrock, manage your Bedrock account or build knowledge bases for you. If your goal is to call Claude models from an agent, the separate Anthropic server is described as doing that, with tools to send a message, list models and count tokens.

It is also not a general AWS control surface. The AWS server is listed with 7 tools covering S3 buckets and objects, EC2 instances, Lambda invocation, DynamoDB queries and CloudWatch logs. That is a different job from asking a knowledge base a question, and the two are not interchangeable. The AWS listing also says a published package or reachable endpoint could not be verified, so no install command is shown there, and no CheckMCP audit has been published for it.

Finally, retrieval is only as good as the knowledge base behind it. If the content was never indexed, is out of date or is poorly organised, the agent will return thin or stale answers. The server gives the agent access. It does not improve the underlying material.

What IAM permissions and costs should you watch?

Treat the credentials you give this server as the agent's own. Whatever the AWS identity behind it can do, the agent can attempt. Create a dedicated identity for the server, grant it only what retrieval from the intended knowledge base needs, and avoid reusing an administrator profile. The repository's README is the place to confirm exactly which actions the server calls. For the general principle, see MCP Server Permissions and Scopes Explained.

On cost, the listing states no price for the server itself, and the repository is Apache-2.0 licensed. Charges for the underlying AWS services are a separate matter and are set by AWS, so check the AWS pricing pages for your account and region before you let an agent query freely. An agent that loops over many questions can generate many retrievals, so watch usage in your billing view, especially in the first days.

Before installing, it is worth auditing any server you are about to connect to company data. How to Audit an MCP Server Before Install lists the checks, such as reading the licence, the maintenance history and the documentation, that the CheckMCP audit already summarises for this repository.

Where is the AWS Bedrock MCP listing?

The full entry is at AWS Bedrock KB. It carries the install commands, the supported clients, the CheckMCP audit summary with a link to the complete report, and a link to the GitHub repository. It also offers an optional Listed on badge for maintainers to paste into a README, which changes nothing about the listing itself.

If your needs are broader than knowledge-base retrieval, compare it with the AWS listing for storage, compute and logs, or the Anthropic listing for calling Claude models. The three do different jobs, and many setups would pick one of them rather than all three.

For the wider picture of how agents use servers like these, start with What Is MCP (Model Context Protocol)? Explained. And if you want a deployment pattern that touches AWS, the Container Deployment to AWS recipe shows one way to combine servers around it.