Query AWS Cost Explorer, Budgets, and Compute Optimizer through a single natural language chat interface. Eliminate manual navigation across multiple consoles to analyze your cloud spend.
Overview
This Guidance shows how to simplify AWS cost management across multiple accounts by building a conversational AI agent that consolidates billing data from multiple sources into a single natural language interface. Finance teams can ask questions like "What are my top compute cost drivers this month?" and the agent queries AWS Cost Explorer for historical spending, AWS Budgets for budget status, and AWS Compute Optimizer for rightsizing recommendations, then synthesizes the data into an immediate conversational response. The agent maintains conversation context across interactions, enabling follow-up questions without repeating information. You can eliminate manual cost report generation, accelerate budget analysis, identify optimization opportunities faster, and empower non-technical finance teams to access complex billing data through simple conversations.
Benefits
Consolidate cost insights through conversation
Democratize FinOps across your organization
Enable non-technical stakeholders to access cost analysis and optimization recommendations without specialized cloud billing expertise. Reduce dependency on dedicated FinOps engineers for routine cost inquiries.
Accelerate analysis with multi-turn context
Conduct complex cost investigations across follow-up questions without repeating prior context. Resolve multi-step FinOps queries in minutes instead of hours of manual cross-referencing.
How it works
This architecture diagram shows how to build a conversational FinOps agent that consolidates AWS cost data using Amazon Bedrock AgentCore, MCP servers, and natural language.
Download the architecture diagram.
Step 1
Administrative users use AWS CDK to deploy the guidance with a single script, uploading application code to Amazon S3 bucket and triggering AWS CodeBuild to build container images stored in Amazon Elastic Container Registry (ECR) for the Amazon Bedrock AgentCore runtime.
Users access the web application hosted on AWS Management Portal for vCenter, which serves the frontend interface.
Users authenticate with Amazon Cognito. Amazon Cognito validates your credentials and returns temporary AWS credentials from the Identity Pool.
The frontend sends the user's question to the Amazon Bedrock AgentCore Runtime — a secure, serverless environment that hosts and runs the agent with session isolation — using the temporary AWS credentials to call InvokeAgentRuntime via IAM SigV4 authentication.
The Strands agent — an open-source agent framework supported natively by AgentCore Runtime — sends the user's question with 24 tool definitions to Claude Sonnet 4.5 on Amazon Bedrock, a fully managed service providing secure access to foundation models. The model selects the appropriate cost analysis tool.
Amazon Bedrock AgentCore Memory — a fully managed service for session and long-term memory — maintains conversation context across interactions, enabling the agent to understand follow-up questions and provide coherent multi-turn cost analysis without users repeating context.
The agent routes the tool call to AgentCore Gateway using IAM SigV4 authentication via InvokeGateway.
AgentCore Identity — a secure identity and credential management service purpose-built for AI agents — retrieves an OAuth 2.0 access token from the registered Amazon Cognito M2M credential provider (using the client credentials grant) and attaches it to the outbound MCP request, enabling the agent to securely access the billing tools.
The Gateway sends the Model Context Protocol (MCP) tool call request with the OAuth token to the Billing MCP Runtime.
The Billing MCP Runtime queries the appropriate AWS cost services: AWS Cost Explorer for historical cost and usage data, AWS Budgets for budget status and alerts, AWS Compute Optimizer for rightsizing recommendations, and AWS Cost & Pricing APIs for current service pricing — providing comprehensive FinOps coverage through a single conversational interface.
Cost data flows back through the chain. The agent sends it to Amazon Bedrock, where Claude generates a natural language summary of your costs.
The formatted response displays the cost breakdown in the user's chat interface.
Amazon CloudWatch provides centralized monitoring, logging, and alerting across all guidance services for complete observability.
Deploy with confidence
Everything you need to launch this Guidance in your account is right here.
Let's make it happen
Ready to deploy? Review the sample code on GitHub for detailed deployment instructions to deploy as-is or customize to fit your needs.
Related content
Build a FinOps agent using Amazon Bedrock AgentCore
Learn how to build a conversational FinOps agent using Amazon Bedrock AgentCore.