AWS open source MCP server for Bedrock Agent to simplify AI agent development

AWS released the Amazon Bedrock Agent’s Open Source Model Context Protocol (MCP) server, providing a direct path from natural language prompts in the proxy IDE to the AgentCore runtime that can deploy the proxy. The package features automatic conversion, environment configuration, and gateway/tool ​​hooks, designed to push typical multi-step integration work into conversation commands.

So, what exactly is it?

The “AgentCore MCP Server” exposes task-specific tools to clients (e.g., Kiro, Claude Code, Cursor, Amazon Q Developer CLI or VS Code Q plug-in) and directs the assistant: (1) minimize existing agents for existing agents for the agent runtime model; (2) configure and configure the AWS environment (credentials, roles/permissions, ECR, configuration files); (3) connect the proxy circuit gateway to the tool call; (4) call and test deployed agents – from the chat surface of the IDE.

In fact, the server teaches your coding assistant to convert entry points into AgentCore handler, please add bedrock_agentcore Import, generate requirements.txtand then rewrite the proxy directly to the payload-based handler that is compatible with the runtime. It can then call the AgentCore CLI to deploy and exercise the agent, including end-to-end calls through the gateway tool.

How to install it? What is customer support?

AWS uses a lightweight launcher (uvx) and standards mcp.json The input and output of most clients with MCP. The AWS team lists what to expect mcp.json Kiro’s location (.kiro/settings/mcp.json), cursor(.cursor/mcp.json), Amazon Q CLI (~/.aws/amazonq/mcp.json) and claude code (~/.claude/mcp.json).

The repository is located in AWSLABS “MCP” Mono-Repo (license 2.0). Root Repo also links to a wider range of AWS MCP resources and documentation when the AgentCore server directory is hosted.

Architecture guidance and “hierarchical” context model

AWS recommends a layered approach to providing the context of the IDE’s assistants growing richer: starting with the proxy client, then adding the AWS document MCP server, a layer in the framework document (e.g., Strands Agents, Langgraph), including the AgentCore and Agent-Framework SDK documents, and ultimately booting the workflow through Per-Zide through Per-Zide “Steering” “Steering” “Steering” “Steering” “Steering” “Steering” “Steering” “Steering” “Steering” “Steering” “Steering Files”. This arrangement reduces retrieval errors and helps the assistant plan end-to-end transformation/deployment/test loops without manual context switching.

Development workflow (typical path)

1. Bootstrap: Use local tools or MCP server. Either provide LAMBDA targets for the AgentCore gateway or deploy the server directly to the AgentCore runtime.
2. Author/Reconstruction: Start with Strands Agents or Langgraph code. The server instructs the Assistant to runtime compatibility handlers, imports, and dependencies to convert.
3. deploy: The assistant finds relevant documents and calls the AgentCore CLI to deploy.
4. Testing and iteration: Call the agent through natural language; if you need tools, please integrate the gateway (MCP client in the agent), re-department (V2) and re-test.

What’s the difference?

Most “agent frameworks” still require developers to learn cloud-specific runtime, credentials, role policies, registry, and deployment CLI before any useful iteration. AWS’s MCP servers are moved to the IDE Assistant and close the “rapid production” gap. Since it is just another MCP server, it consists of an existing DOC server (AWS Service Documentation, Strands, Langgraph) and can be improved on the client side of MCP Aware Aware, which makes it a low friction entry point for teams of standardized base agents.

I like that AWS provides a real MCP endpoint for AgentCore that my IDE can call directly. this uvx-based on mcp.json Config enables the client to connect trivia (cursor, Claude Code, Kiro, Amazon Q Cli) and server tools to clearly map to the AgentCore Runtime/Gateway/Memory Stack, while retaining the existing strands/langgraph code path. In fact, this will prompt → Refactor → Deploy → Test loops fold into reproducible, scriptable workflows instead of custom glue code.

Check Github repository and Technical details. Check out ours anytime Tutorials, codes and notebooks for github pages. Also, please stay tuned for us twitter And don’t forget to join us 100K+ ml reddit And subscribe Our newsletter. wait! Are you on the telegram? Now, you can also join us on Telegram.

Asif Razzaq is CEO of Marktechpost Media Inc. As a visionary entrepreneur and engineer, ASIF is committed to harnessing the potential of artificial intelligence to achieve social benefits. His recent effort is to launch Marktechpost, an artificial intelligence media platform that has an in-depth coverage of machine learning and deep learning news that can sound both technically, both through technical voices and be understood by a wide audience. The platform has over 2 million views per month, demonstrating its popularity among its audience.