AWS MCP Server Reaches General Availability: Secure, Up-to-Date AWS Access for AI Agents

Introduction

The AWS MCP (Model Context Protocol) Server has officially moved from preview to general availability, marking a significant milestone for developers building AI agents and coding assistants on Amazon Web Services. This managed remote server provides a streamlined, secure bridge for AI agents to interact with AWS services using authenticated access—without exposing full administrative privileges. As part of the Agent Toolkit for AWS, the MCP Server joins a suite of skills and plugins designed to help AI agents build more effectively on AWS.

The Challenge: AI Agents and AWS Complexity

AI coding agents have proven valuable for automating repetitive tasks, but they face real obstacles when working with AWS at scale. Without direct access to current AWS documentation, agents often rely on outdated training data—potentially missing new services like Amazon S3 Vectors, Aurora DSQL, or Bedrock AgentCore. Moreover, when asked to provision infrastructure, agents typically default to the AWS CLI rather than using infrastructure-as-code tools like AWS CDK or CloudFormation. They also tend to generate overly permissive IAM policies that work in demos but fail production security audits. The result is infrastructure that is brittle and not ready for real-world use.

How the AWS MCP Server Addresses These Issues

The AWS MCP Server provides a compact, fixed set of tools that do not consume the AI model’s context window—a critical limitation in many agents. Key tools include:

• call_aws: Executes any of over 15,000 AWS API operations using the user’s existing IAM credentials. New APIs are supported within days of launch.
• search_documentation and read_documentation: Retrieve the latest AWS documentation and best practices in real time, ensuring the agent always works from current information rather than stale training data.

By integrating these tools, the MCP Server lets agents use proper AWS SDKs, adopt secure IAM practices, and reference up-to-date guidance—tackling the core problems head-on.

New Capabilities in General Availability

With the general availability release, several enhancements improve security, performance, and ease of use:

• IAM context keys: Previously required a separate permission to use the server; now you can express fine-grained access control directly in standard IAM policies. This simplifies setup and aligns with existing security practices.
• No authentication for documentation retrieval: Agents can fetch AWS documentation without needing additional credentials, reducing friction for read-only queries.
• Reduced token consumption: The server now uses fewer tokens per interaction, which is especially beneficial for complex, multi-step workflows where context limits are a bottleneck.

These updates make the server more efficient and easier to integrate into existing agent pipelines while maintaining strong security boundaries.

The run_script Tool: Sandboxed Python Execution

Perhaps the most notable new feature is the run_script tool. This allows an AI agent to write and execute a short Python script server-side in a sandboxed environment. The sandbox inherits the user’s IAM permissions but has no network access, preventing the script from interacting with local files or spawning arbitrary shell commands.

The practical benefit is significant: when an agent needs to call multiple APIs and combine results, executing individual API calls sequentially is slow and consumes valuable context. With run_script, the agent can chain API calls, filter responses, and compute the final result in a single round trip—dramatically faster and more token-efficient. This opens up new possibilities for complex data processing and orchestration tasks within agent workflows.

From SOPs to Skills: Curated Guidance for Agents

Another important shift is the transition from Agent SOPs (Standard Operating Procedures) to Skills. Skills provide curated, context-aware guidance and best practices for common AWS tasks. Whereas SOPs were static, Skills are designed to evolve, offering the agent up-to-date, actionable recommendations for activities such as setting up a VPC, deploying a serverless application, or configuring IAM roles. This change makes it easier for agents to follow AWS-recommended practices from the start, further improving the quality of the generated infrastructure.

Conclusion

The general availability of the AWS MCP Server delivers a mature, secure, and efficient way for AI agents to interact with AWS. By addressing key pain points—such as outdated documentation, improper tool selection, and overly broad IAM policies—the server helps developers build production-ready infrastructure with confidence. With new capabilities like IAM context keys, reduced token usage, the run_script sandbox, and the shift to Skills, the AWS MCP Server is a valuable addition to the Agent Toolkit. Start exploring today to see how it can enhance your AI-driven workflows on AWS.