AWS Cost Explorer and Amazon Bedrock Model Invocation Logs MCP Server & Client

What is AWS Cost Explorer and Amazon Bedrock Model Invocation Logs MCP Server & Client?

This MCP server retrieves AWS spend data via Cost Explorer and Amazon Bedrock usage data from CloudWatch Model invocation logs. It integrates with Anthropic’s Model Context Protocol (MCP) so users can analyze cloud costs and Bedrock usage through natural language conversations with Claude.

How to use AWS Cost Explorer and Amazon Bedrock Model Invocation Logs MCP Server & Client?

Install uv, clone the repository, create a Python 3.12 virtual environment, and install dependencies. Set your AWS credentials, export MCP_TRANSPORT (use stdio for local or sse for remote) and BEDROCK_LOG_GROUP_NAME, then run python server.py. Configure Claude Desktop by adding a Docker or UV entry to claude_desktop_config.json. For a remote server, deploy on EC2 with MCP_TRANSPORT=sse and test using mcp_sse_client.py.

Key features of AWS Cost Explorer and Amazon Bedrock Model Invocation Logs MCP Server & Client

• Amazon EC2 spend analysis for the last day.
• Amazon Bedrock spend analysis by region, user, and model.
• Service spend reports across all AWS services.
• Detailed cost breakdown by day, region, service, and instance type.
• Interactive natural language interface via Claude.

Use cases of AWS Cost Explorer and Amazon Bedrock Model Invocation Logs MCP Server & Client

• Ask “What was my EC2 spend yesterday?” and get an instant breakdown.
• Generate a weekly Bedrock usage report per model and user.
• Identify top five AWS services by cost over the past month.
• Analyze spending by region for the last 14 days.
• Build a LangGraph agent that answers cost questions using remote MCP tools.

FAQ from AWS Cost Explorer and Amazon Bedrock Model Invocation Logs MCP Server & Client

What are the requirements to run this server?

Python 3.12, AWS credentials with read-only access to Cost Explorer and CloudWatch, and Anthropic API access for Claude integration. Optionally, Bedrock access for the LangGraph agent and EC2 for a remote server.

How do I configure Claude Desktop?

Add either a Docker entry (using the aws-cost-explorer-mcp:latest image) or a UV entry pointing to your local repository in the mcpServers section of claude_desktop_config.json. Include your AWS credentials and Bedrock log group name as environment variables.

What tools does the server expose to Claude?

Four tools: get_ec2_spend_last_day(), get_detailed_breakdown_by_day(days), get_bedrock_daily_usage_stats(days, region, log_group_name), and get_bedrock_hourly_usage_stats(days, region, log_group_name).

Does this server work with remote MCP clients?

Yes. Set MCP_TRANSPORT=sse to run the server on EC2. Claude Desktop does not currently support remote MCP servers, but you can test with the provided mcp_sse_client.py or use the Chainlit app with a LangGraph agent.

Is the MCP protocol authenticated or encrypted?

No. MCP uses JSON-RPC 2.0 without built-in authentication or authorization. Avoid sending sensitive data. For secure remote access, the README describes an optional nginx reverse proxy with HTTPS using an SSL certificate.