AWS has officially declared general availability (GA) for its AWS Knowledge MCP Server, ushering in the next-gen way AI agents talk to AWS.
MCP (Model Context Protocol) is rapidly emerging as a standard for connecting LLMs and agent frameworks to external “context providers” (think: documentation, API state, best practices). AWS is now playing in the MCP space with a set of servers tailored for AWS domains.
The Knowledge MCP Server is one of these: a remote, fully managed service that gives your AI assistants live access to AWS documentation, blog posts, new-feature announcements, Well-Architected guidance, and even regional API/CloudFormation availability details.
Why developers should care
AWS have done a great job on documenting many of their core services right out of the bat.
- Instant accuracy, less hallucination – Instead of hoping your LLM “knew” the latest on S3 or Lambda, your assistant can refer to the canonical AWS source.
- No setup, no creds– The server is publicly accessible, with no AWS account needed (though it’s rate-limited).
- Global and always current – AWS handles the content updates. You just hook your MCP client into
https://knowledge-mcp.global.api.aws. - Composable architecture – The “Knowledge” server is just one piece. Other AWS MCP servers (for APIs, pricing, etc.) plug into the same protocol space.
How to get started
There’s a lot more to learn and understand in the official documentation.
- Ensure your MCP client supports remote servers (or use a proxy like
fastmcp) - Add a block in your MCP config, e.g.:
{ "mcpServers": { "aws-knowledge-mcp-server": { "url": "https://knowledge-mcp.global.api.aws" } } } - From your AI agent or chat tool, ask things like:
- “Search AWS docs for DynamoDB partition key best practices”
- “Which AWS regions currently support X resource type?”
- “What’s new in AWS SimplyDeploy this month?”
You can also use the MCP Inspector tool to make calls directly.
Big picture: What AWS is doing with MCP
This move cements AWS’s intent to make cloud development more conversational and less error-prone. With MCP servers, the barrier between “I want to know how X works” and “tell me what to code” gets thinner.
Earlier this year, AWS released open-source MCP servers for ECS, EKS, serverless domains, etc. The Knowledge MCP complements those by focusing on documentation and advice, not just API execution.
Expect more MCP servers (e.g. for cost, security, monitoring) in the months ahead.
