Это русская версия материала. Для полноты языковых маршрутов текст основан на существующей основной версии.
AWS MCP Server is officially GA.
If you are still managing cloud resources through the old loop of configuring AWS CLI, writing scripts, setting permissions, and running automation, this update is worth a look.
What AWS MCP Server Is
MCP, short for Model Context Protocol, is the open protocol Anthropic introduced and later moved toward broader industry standardization through the Linux Foundation. In plain English: MCP is a standard interface for AI agents to talk to external tools.
AWS MCP Server exposes core AWS cloud services, including EC2, S3, RDS, and Lambda, through MCP. Any MCP-compatible agent, such as Claude Desktop, Cursor, or another MCP client, can use natural language to manage AWS resources.
What It Can Do
According to AWS documentation, AWS MCP Server supports operations such as:
- Querying resource status: checking EC2 instance state, S3 buckets, and RDS database configuration
- Running operations tasks: restarting instances, scaling resources, and creating backups
- Checking costs: reviewing current bills and usage trends
- Managing security groups: changing inbound and outbound rules
These workflows used to require the AWS Console, CLI, or Infrastructure-as-Code tooling. Now an agent can understand the intent and execute the operation directly.
Practical Availability
GA means AWS considers the service stable enough for production use. A few boundaries still matter:
- Permissions: MCP Server requires IAM roles and policies. The agent only gets the permissions you grant it.
- Coverage: the first wave focuses on core services, not all 200+ AWS services.
- Audit logs: operations executed through MCP Server are recorded in CloudTrail.
Industry Signal
AWS is not the first cloud vendor to build an MCP server. Google Cloud and Azure are moving in the same direction. But AWS reaching GA is a signal: MCP is moving from experimental protocol to standard cloud infrastructure interface.
For developers, that means less custom agent integration code for every cloud vendor. Write against MCP once, and at least in theory, reuse that pattern across multiple platforms.
Of course, "in theory" is doing work here. Each cloud has its own service model and permission system. Cross-cloud reuse will not be trivial. But the direction is right.
Main sources:
- AWS MCP Server GA announcement
- AWS MCP Server documentation