mcp/aws-pricing-mcp-server repository overview

{
  // Required parameters
  "pricing_data": {
    // This should contain pricing data retrieved from get_pricing
    "status": "success",
    "service_name": "bedrock",
    "data": "... pricing information ...",
    "message": "Retrieved pricing for bedrock from AWS Pricing url"
  },
  "service_name": "Amazon Bedrock",

  // Core parameters (commonly used)
  "related_services": ["Lambda", "S3"],
  "pricing_model": "ON DEMAND",
  "assumptions": [
    "Standard ON DEMAND pricing model",
    "No caching or optimization applied",
    "Average request size of 4KB"
  ],
  "exclusions": [
    "Data transfer costs between regions",
    "Custom model training costs",
    "Development and maintenance costs"
  ],
  "output_file": "cost_analysis_report.md",  // or "cost_analysis_report.csv" for CSV format
  "format": "markdown",  // or "csv" for CSV format

  // Advanced parameter for complex scenarios
  "detailed_cost_data": {
    "services": {
      "Amazon Bedrock Foundation Models": {
        "usage": "Processing 1M input tokens and 500K output tokens with Claude 3.5 Haiku",
        "estimated_cost": "$80.00",
        "free_tier_info": "No free tier for Bedrock foundation models",
        "unit_pricing": {
          "input_tokens": "$0.0008 per 1K tokens",
          "output_tokens": "$0.0016 per 1K tokens"
        },
        "usage_quantities": {
          "input_tokens": "1,000,000 tokens",
          "output_tokens": "500,000 tokens"
        },
        "calculation_details": "$0.0008/1K × 1,000K input tokens + $0.0016/1K × 500K output tokens = $80.00"
      },
      "AWS Lambda": {
        "usage": "6,000 requests per month with 512 MB memory",
        "estimated_cost": "$0.38",
        "free_tier_info": "First 12 months: 1M requests/month free",
        "unit_pricing": {
          "requests": "$0.20 per 1M requests",
          "compute": "$0.0000166667 per GB-second"
        },
        "usage_quantities": {
          "requests": "6,000 requests",
          "compute": "6,000 requests × 1s × 0.5GB = 3,000 GB-seconds"
        },
        "calculation_details": "$0.20/1M × 0.006M requests + $0.0000166667 × 3,000 GB-seconds = $0.38"
      }
    }
  },

  // Recommendations parameter - can be provided directly or generated
  "recommendations": {
    "immediate": [
      "Optimize prompt engineering to reduce token usage for Claude 3.5 Haiku",
      "Configure Knowledge Base OCUs based on actual query patterns",
      "Implement response caching for common queries to reduce token usage"
    ],
    "best_practices": [
      "Monitor OCU utilization metrics and adjust capacity as needed",
      "Use prompt caching for repeated context across API calls",
      "Consider provisioned throughput for predictable workloads"
    ]
  }
}

**PURPOSE:** Access complete AWS pricing datasets as downloadable files for historical analysis and bulk processing.

**WORKFLOW:** Use this for historical pricing analysis or bulk data processing when current pricing from get_pricing() isn't sufficient.

**PARAMETERS:**
- Service code from get_pricing_service_codes() (e.g., 'AmazonEC2', 'AmazonS3')
- AWS region (e.g., 'us-east-1', 'eu-west-1')
- Optional: effective_date for historical pricing (default: current date)

**RETURNS:** Dictionary with download URLs for different formats:
- 'csv': Direct download URL for CSV format
- 'json': Direct download URL for JSON format

**USE CASES:**
- Historical pricing analysis (get_pricing() only provides current pricing)
- Bulk data processing without repeated API calls
- Offline analysis of complete pricing datasets
- Savings Plans analysis across services

**FILE PROCESSING:**
- CSV files: Lines 1-5 are metadata, Line 6 contains headers, Line 7+ contains pricing data
- Use `tail -n +7 pricing.csv | grep "t3.medium"` to filter data

**PARAMETERS:**
- service_code (required): AWS service code (e.g., 'AmazonEC2', 'AmazonS3', 'AmazonES')
- region (optional): AWS region string (e.g., 'us-east-1') OR list for multi-region comparison (e.g., ['us-east-1', 'eu-west-1']). Omit for global services like DataTransfer or CloudFront that don't have region-specific pricing.
- filters (optional): List of filter dictionaries in format {'Field': str, 'Type': str, 'Value': str}
- max_allowed_characters (optional): Response size limit in characters (default: 100,000, use -1 for unlimited)
- output_options (optional): OutputOptions object for response transformation and size reduction
- max_results (optional): Maximum number of results to return per page (default: 100, min: 1, max: 100)
- next_token (optional): Pagination token from previous response to get next page of results

**MANDATORY WORKFLOW - ALWAYS FOLLOW:**

**Step 1: Discover Available Options**
```python
service_codes = get_pricing_service_codes()                              # Find correct service (skip if known)
attributes = get_pricing_service_attributes('AmazonEC2')                 # Discover filterable dimensions
attribute_values = get_pricing_attribute_values('AmazonEC2', 'memory')   # Get valid values for filtering
```

**Step 2: Build Precise Filters**
```python
# Use ONLY values discovered in Step 1
filters = [
   {"Field": "memory", "Value": ["8 GiB", "16 GiB", "32 GiB"], "Type": "ANY_OF"},     # Multiple options
   {"Field": "instanceType", "Value": "m5", "Type": "CONTAINS"},                      # Pattern matching
   {"Field": "instanceType", "Value": ["t2", "m4"], "Type": "NONE_OF"}                # Exclude older

**Step 3: Execute Query**
```python
pricing = get_pricing('AmazonEC2', 'us-east-1', filters)
```

**FILTER TYPES:**
- **EQUALS**: Exact match (default) - `{"Field": "instanceType", "Value": "m5.large"}`
- **ANY_OF**: Multiple options - `{"Field": "memory", "Value": ["8 GiB", "16 GiB"], "Type": "ANY_OF"}`
- **CONTAINS**: Pattern match - `{"Field": "instanceType", "Value": "m5", "Type": "CONTAINS"}`
- **NONE_OF**: Exclusion - `{"Field": "instanceType", "Value": ["t2", "m4"], "Type": "NONE_OF"}`

**CRITICAL: ANY_OF FILTER VALUE LIMITS:**
- **1024 CHARACTER LIMIT**: Total length of all values in ANY_OF arrays cannot exceed 1024 characters
- **PROGRESSIVE FILTERING**: Start with minimal qualifying options, expand if needed
- **EXAMPLE VIOLATION**: `["8 GiB", "16 GiB", "32 GiB", "64 GiB", "96 GiB", "128 GiB", ...]` (TOO LONG)
- **CORRECT APPROACH**: `["8 GiB", "16 GiB", "32 GiB", "36 GiB", "48 GiB"]` (TARGETED LIST)

**COMMON USE CASES:**

**COST OPTIMIZATION - EXHAUSTIVE MINIMUM-FIRST APPROACH:** When users ask for "lowest price", "cheapest", or cost optimization
- **LOWER = CHEAPER ASSUMPTION**: For cost optimization, assume lower capabilities cost less than higher ones
  * 32 GB storage is cheaper than 300 GB storage
  * 8 GiB RAM is cheaper than 64 GiB RAM
- **CRITICAL FOR COST QUERIES**: Start IMMEDIATELY above minimum requirement and test ALL options incrementally
- **EXHAUSTIVE ENUMERATION REQUIRED**: Each storage/memory tier is MUTUALLY EXCLUSIVE - must list each one explicitly
- **STOP AT REASONABLE UPPER BOUND**: For cost optimization, limit upper bound to 2-3x minimum requirement to avoid expensive options
- **exclude_free_products**: ESSENTIAL for cost analysis - removes $0.00 reservation placeholders, SQL licensing variants, and special pricing entries that obscure actual billable instances when finding cheapest options
- Use ANY_OF for efficient multi-option comparison in single API call
- Multi-attribute capability filtering for minimum requirements
- Combine CONTAINS + NONE_OF for refined discovery

**OUTPUT OPTIONS (Response Size & Performance Control):**
- **PURPOSE**: Transform and optimize API responses for ALL services, especially critical for large services (EC2, RDS)
- **IMMEDIATE COMBINED APPROACH**: `{"pricing_terms": ["OnDemand", "FlatRate"], "product_attributes": ["instanceType", "location", "memory"]}`
- **ATTRIBUTE DISCOVERY**: Use get_pricing_service_attributes() - same names for filters and output_options
- **SIZE REDUCTION**: 80%+ reduction with combined pricing_terms + product_attributes
- **exclude_free_products**: Remove products with $0.00 OnDemand pricing (useful when you know service has paid tiers)
- **WHEN TO USE**: Always for large services, recommended for all services to improve performance

**CRITICAL REQUIREMENTS:**
- **NEVER GUESS VALUES**: Always use get_pricing_attribute_values() to discover valid options
- **EXHAUSTIVE ENUMERATION**: For cost optimization, list ALL qualifying tiers individually - they are mutually exclusive
- **USE SPECIFIC FILTERS**: Large services (EC2, RDS) require 2-3 filters minimum
- **NEVER USE MULTIPLE CALLS**: When ANY_OF can handle it in one call
- **VERIFY EXISTENCE**: Ensure all filter values exist in the service before querying
- **FOR "CHEAPEST" QUERIES**: Focus on lower-end options that meet minimum requirements, test incrementally
- **EXPLORE ALTERNATIVES**: When response includes "alternatives" field, MUST fetch their pricing if applicable to the use case before answering

**CONSTRAINTS:**
- **CURRENT PRICING ONLY**: Use get_price_list_urls for historical data
- **NO SPOT/SAVINGS PLANS**: Only OnDemand, FlatRate, and Reserved Instance pricing available (ANY combination possible)
- **CHARACTER LIMIT**: 100,000 characters default response limit (use output_options to reduce)
- **REGION AUTO-FILTER**: Region parameter automatically creates regionCode filter

**ANTI-PATTERNS:**
- DO NOT make multiple API calls that could be combined with ANY_OF
- DO NOT build cross-products manually when API can handle combinations
- DO NOT call get_pricing_service_codes() when service code is already known (e.g., "AmazonEC2")
- DO NOT use EQUALS without first checking get_pricing_attribute_values()
- DO NOT skip discovery workflow for any use case
- DO NOT use broad queries without specific filters on large services
- DO NOT assume attribute values exist across different services/regions
- DO NOT skip intermediate tiers: Missing 50GB, 59GB options when testing 32GB → 75GB jump
- DO NOT set upper bounds too high: Including 500GB+ storage when user needs ≥30GB (wastes character limit)
- DO NOT ignore alternatives field or use only ["OnDemand"] in output_options

**EXAMPLE USE CASES:**

**1. Cost-Optimized Multi-Attribute Filtering (CORRECT APPROACH):**
```python
# Find cheapest EC2 instances meeting minimum requirements (>= 8 GiB memory, >= 30 GB storage)
# EXHAUSTIVE ENUMERATION of qualifying tiers - each is mutually exclusive
filters = [
   {"Field": "memory", "Value": ["8 GiB", "16 GiB", "32 GiB"], "Type": "ANY_OF"},  # All tiers ≥8GB up to reasonable limit
   {"Field": "storage", "Value": ["1 x 32 SSD", "1 x 60 SSD", "1 x 75 NVMe SSD"], "Type": "ANY_OF"},  # All tiers ≥30GB up to reasonable limit
   {"Field": "instanceType", "Value": ["t2", "m4"], "Type": "NONE_OF"},  # Exclude older generations
   {"Field": "tenancy", "Value": "Shared", "Type": "EQUALS"}  # Exclude more expensive dedicated
]
pricing = get_pricing('AmazonEC2', 'us-east-1', filters)
```

**2. Efficient Multi-Region Comparison:**
```python
# Compare same configuration across regions - use region parameter for multi-region
filters = [{"Field": "instanceType", "Value": "m5.large", "Type": "EQUALS"}]
pricing = get_pricing('AmazonEC2', ['us-east-1', 'us-west-2', 'eu-west-1'], filters)
```

**3. Large service with output optimization (recommended approach):**
```python
output_options = {"pricing_terms": ["OnDemand", "FlatRate"], "product_attributes": ["instanceType", "location"], "exclude_free_products": true}
pricing = get_pricing('AmazonEC2', 'us-east-1', filters, output_options=output_options)
```

**4. Pattern-Based Discovery:**
```python
# Find all Standard storage tiers except expensive ones
filters = [
    {"Field": "storageClass", "Value": "Standard", "Type": "CONTAINS"},
    {"Field": "storageClass", "Value": ["Standard-IA"], "Type": "NONE_OF"}
]
```

**FILTERING STRATEGY:**
- **Large Services (EC2, RDS)**: ALWAYS use 2-3 specific filters to prevent 200+ record responses
- **Small Services**: May work with single filter or no filters
- **Multi-Option Analysis**: Use ANY_OF instead of multiple API calls
- **Pattern Discovery**: Use CONTAINS for finding families or tiers
- **Smart Exclusion**: Use NONE_OF for compliance or cost filtering

**SUCCESS CRITERIA:**
- Used discovery workflow (skip get_pricing_service_codes() if service known)
- Applied appropriate filters for the service size
- Used exact values from get_pricing_attribute_values()
- Used ANY_OF for multi-option scenarios instead of multiple calls
- For cost optimization: tested ALL qualifying tiers exhaustively (in a reasonable range)
- Included ["OnDemand", "FlatRate"] in output_options and explored all alternatives

**PURPOSE:** Discover what values are available for specific pricing filter attributes of an AWS service.

**WORKFLOW:** Use this after get_pricing_service_attributes() to see valid values for each filter attribute.

**PARAMETERS:**
- Service code from get_pricing_service_codes() (e.g., 'AmazonEC2', 'AmazonRDS')
- List of attribute names from get_pricing_service_attributes() (e.g., ['instanceType', 'location'])
- filters (optional): Dictionary mapping attribute names to regex patterns (e.g., {'instanceType': 't3'})

**RETURNS:** Dictionary mapping attribute names to their valid values. Filtered attributes return only matching values, unfiltered attributes return all values.

**EXAMPLE RETURN:**
```
{
    'instanceType': ['t2.micro', 't3.medium', 'm5.large', ...],
    'location': ['US East (N. Virginia)', 'EU (London)', ...]
}
```

**NEXT STEPS:** Use these values in get_pricing() filters to get specific pricing data.

**ERROR HANDLING:** Uses "all-or-nothing" approach - if any attribute fails, the entire operation fails.

**EXAMPLES:**
- Single attribute: ['instanceType'] returns {'instanceType': ['t2.micro', 't3.medium', ...]}
- Multiple attributes: ['instanceType', 'location'] returns both mappings
- Partial filtering: filters={'instanceType': 't3'} applies only to instanceType, location returns all values

**PURPOSE:** Discover what pricing dimensions (filters) are available for a specific AWS service.

**WORKFLOW:** Use this after get_pricing_service_codes() to see what filters you can apply to narrow down pricing queries.

**PARAMETERS:**
- service_code: AWS service code from get_pricing_service_codes() (e.g., 'AmazonEC2', 'AmazonRDS')
- filter (optional): Case-insensitive regex pattern to filter attribute names (e.g., "instance" matches "instanceType", "instanceFamily")

**RETURNS:** List of attribute names (e.g., 'instanceType', 'location', 'storageClass') that can be used as filters.

**NEXT STEPS:**
- Use get_pricing_attribute_values() to see valid values for each attribute
- Use these attributes in get_pricing() filters to get specific pricing data

**EXAMPLE:** For 'AmazonRDS' you might get ['engineCode', 'instanceType', 'deploymentOption', 'location'].

**PURPOSE:** Discover which AWS services have pricing information available in the AWS Price List API.

**PARAMETERS:**
- filter (optional): Case-insensitive regex pattern to filter service codes (e.g., "bedrock" matches "AmazonBedrock", "AmazonBedrockService")

**WORKFLOW:** This is the starting point for any pricing query. Use this first to find the correct service code.

**RETURNS:** List of service codes (e.g., 'AmazonEC2', 'AmazonS3', 'AWSLambda') that can be used with other pricing tools.

**NEXT STEPS:**
- Use get_pricing_service_attributes() to see what filters are available for a service
- Use get_pricing() to get actual pricing data for a service

**NOTE:** Service codes may differ from AWS console names (e.g., 'AmazonES' for OpenSearch, 'AWSLambda' for Lambda).

{
  "mcpServers": {
    "aws-pricing": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "AWS_REGION",
        "-e",
        "AWS_PROFILE",
        "-e",
        "AWS_ACCESS_KEY_ID",
        "-e",
        "AWS_SECRET_ACCESS_KEY",
        "-e",
        "AWS_SESSION_TOKEN",
        "mcp/aws-pricing-mcp-server"
      ],
      "env": {
        "AWS_REGION": "us-east-1",
        "AWS_PROFILE": "default",
        "AWS_ACCESS_KEY_ID": "AKIAIOSFODNN7EXAMPLE",
        "AWS_SECRET_ACCESS_KEY": "your-aws-secret-access-key",
        "AWS_SESSION_TOKEN": "your-aws-session-token"
      }
    }
  }
}