AWS MCP Server Is Now GA, and AI Agents Just Got Safer
AWS has officially made its managed AWS MCP Server generally available, and the implications for AI-powered cloud development are significant. The server gives AI coding agents controlled, auditable access to AWS APIs, documentation, and operational workflows, all through a standard interface.
The problem it solves is real. AI agents working with AWS have long relied on outdated training data, sometimes months behind. Without current documentation, agents miss services like Amazon S3 Vectors, Amazon Aurora DSQL, or Amazon Bedrock AgentCore. This release changes that directly.
Since its preview at re:Invent 2025, AWS has notably expanded coverage. It now supports all AWS APIs, including long-running operations and file uploads. Furthermore, AWS added sandboxed Python execution for multi-step tasks, meaning agents run code without touching the local filesystem or shell. That is a critical security boundary.
Governance is built in from the start. The server includes IAM-based access controls, CloudWatch metrics, and CloudTrail logging. As a result, organizations audit every agent action independently. Documentation search and skill discovery also work without requiring AWS credentials, lowering the barrier to entry.
The server is now part of the newly announced Agent Toolkit for AWS. This open source toolkit bundles tools, plugins, and workflows that help agents work more reliably with AWS services. It cuts errors, reduces retries, and trims token usage. Sébastien Stormacq, principal developer advocate at AWS, describes the shift plainly: combining current documentation, authenticated API access, and sandboxed execution in one server fundamentally changes what an agent can do on AWS.
The toolkit works with any MCP-compatible agent, including Claude Code, Kiro, Cursor, and Codex. Authentication uses IAM and SigV4. Because the server currently supports OAuth 2.1 only, local credentials pass through the open-source MCP Proxy for AWS, which translates IAM authentication into OAuth-compatible requests.
Not everyone is fully convinced yet. Some practitioners question the value proposition of the server, while others raise security concerns. Kunal Parsewar, a DevOps engineer at ReliaQuest, notes that no gateways currently restrict certain agent operations. That gap could matter at scale.
Still, others see a smart strategic move. Jeremy Daly, founder of Ampt, writes that AWS is pushing hard to become the default platform for AI coding agents, and giving developers an opinionated, authenticated entry point is the right play.
Currently, the service runs in two regions: Northern Virginia and Frankfurt. It is free to use, though charges apply to resources the agents consume.
