CloudWatch MCP Server

This simplified MCP server provides a streamlined way to interact with AWS CloudWatch resources through the MCP protocol. It exposes CloudWatch log groups, log queries, and alarms as resources and tools.

Features

• List all CloudWatch log groups with their metadata

• List all CloudWatch alarms with their current states

• Query CloudWatch logs using CloudWatch Insights across multiple log groups

• Discover available fields across multiple log groups with shared schema

• Automatic JSON parsing for @message field in log queries

• Check if specific log groups exist

• Get detailed information about specific log groups

• Filter alarms by state (all alarms or only those in ALARM state)

• Retrieve all saved CloudWatch Logs Insights queries

Related MCP server: CloudWatch Logs MCP Server

Prerequisites

• Python 3.12 or higher

• AWS credentials configured (via environment variables, AWS CLI, or IAM role)

• MCP CLI (version 0.1.1 or higher)

• Boto3 (AWS SDK for Python)

Setup

1. Make sure you have Python 3.12+ installed.

2. Create a virtual environment (optional but recommended):

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Configure AWS credentials if you haven't already:

   aws configure

   Or set environment variables:

   export AWS_ACCESS_KEY_ID="your-access-key"
   export AWS_SECRET_ACCESS_KEY="your-secret-key"
   export AWS_REGION="your-region"

Project Structure

• cloudwatch_server.py - MCP server implementation for CloudWatch integration

• aws_cloudwatch.py - Simplified AWS CloudWatch integration module

• test_cloudwatch.py - Command-line utility to test the CloudWatch integration

Running the server

Start the MCP server:

python cloudwatch_server.py

Or using the MCP CLI:

mcp run cloudwatch_server.py

Using the MCP server

Resources

The server exposes the following resources:

• cloudwatch://log-groups - Lists all CloudWatch log groups

• cloudwatch://log-groups/{log_group_name} - Gets detailed information about a specific log group

• cloudwatch://alarms - Lists all CloudWatch alarms

• cloudwatch://alarms/in-alarm - Lists only CloudWatch alarms currently in ALARM state

• cloudwatch://saved-queries - Lists all saved CloudWatch Logs Insights queries

Tools

The server provides the following tools:

• query_logs - Query CloudWatch logs using CloudWatch Insights

  • Parameters:

    • log_group_names: Single log group name or list of log group names to query

    • query_string: CloudWatch Insights query string

    • start_time: (Optional) Start time for the query in Unix timestamp milliseconds

    • end_time: (Optional) End time for the query in Unix timestamp milliseconds

  • Features:

    • Automatically parses JSON in @message field

    • Returns structured data for JSON messages

    • Handles multiple log groups in a single query

• discover_log_fields - Discover available fields across multiple log groups

  • Parameters:

    • log_group_names: Single log group name or list of log group names to analyze

  • Features:

    • Efficiently discovers fields across multiple log groups

    • Assumes shared schema across log groups

    • Detects nested JSON fields in @message

    • Identifies field types (number, boolean, string, array)

• log_group_exists - Check if CloudWatch log groups exist

  • Parameters:

    • log_group_names: Single log group name or list of log group names to check

  • Returns:

    • Dictionary mapping each log group to its existence status

• get_saved_queries - Fetch all saved CloudWatch Logs Insights queries

  • No parameters required

Testing the CloudWatch integration

You can test the CloudWatch integration directly using the provided test script:

# Make the test file executable
chmod +x test_cloudwatch.py

# List all log groups
./test_cloudwatch.py log-groups

# List all alarms
./test_cloudwatch.py alarms

# Use a specific AWS profile
./test_cloudwatch.py log-groups --profile my-profile

# Enable verbose logging
./test_cloudwatch.py alarms -v

Examples with MCP CLI

Using the MCP CLI:

# List all log groups
mcp inspect cloudwatch://log-groups

# Get details about a specific log group
mcp inspect cloudwatch://log-groups/my-log-group-name

# List all alarms
mcp inspect cloudwatch://alarms

# List alarms currently in ALARM state
mcp inspect cloudwatch://alarms/in-alarm

# List all saved CloudWatch Logs Insights queries
mcp inspect cloudwatch://saved-queries

# Query logs from multiple log groups using CloudWatch Insights
mcp call query_logs --log_group_names '["log-group-1", "log-group-2"]' --query_string "fields @timestamp, @message | limit 10"

# Query logs from a single log group (still supported)
mcp call query_logs --log_group_names "my-log-group" --query_string "fields @timestamp, @message | limit 10"

# Discover fields across multiple log groups
mcp call discover_log_fields --log_group_names '["log-group-1", "log-group-2"]'

# Check if multiple log groups exist
mcp call log_group_exists --log_group_names '["log-group-1", "log-group-2"]'

# Get all saved CloudWatch Logs Insights queries
mcp call get_saved_queries

License

MIT