Build a FinOps agent using Amazon Bedrock AgentCore
New solution uses Claude Sonnet 4.5 to query AWS cost data across multiple accounts via natural language.
AWS has released a detailed guide for building a specialized FinOps agent using Amazon Bedrock AgentCore, designed to help finance teams manage cloud costs across multiple AWS accounts. The solution consolidates data from AWS Cost Explorer, AWS Budgets, and AWS Compute Optimizer into a single conversational interface powered by Anthropic's Claude Sonnet 4.5 model. It features 30 days of conversation memory, allowing users to ask follow-up questions without repeating context, and integrates over 20 specialized tools covering the full spectrum of cost management from analysis to optimization.
The architecture uses AWS CDK for deployment and consists of several key components: authentication via Amazon Cognito, a frontend hosted on AWS Amplify, and the AgentCore Runtime that processes queries. The system employs Model Context Protocol (MCP) servers to provide access to AWS billing and pricing data, with AgentCore Gateway managing tool invocations. The solution transforms existing AWS Labs MCP servers through a stdio-to-HTTP adaptation, builds ARM64 container images for AWS Graviton processors, and deploys them to AgentCore Runtimes with specific IAM permissions for secure API access.
- Uses Anthropic Claude Sonnet 4.5 with 30-day conversation memory for contextual follow-ups
- Integrates over 20 specialized tools from AWS Cost Explorer, Budgets, and Compute Optimizer
- Built with AWS CDK and uses MCP servers for secure access to billing/pricing APIs
Why It Matters
Democratizes cloud cost management by letting non-technical teams query complex spending data through natural language conversations.