mcp

Exposes 50+ AWS services (Lambda, DynamoDB, S3, CloudWatch, IAM, etc.) as callable tools through the Model Context Protocol, using a unified schema-based function registry that translates MCP tool definitions into AWS SDK calls. Each service gets a dedicated MCP server that implements the MCP specification's tools interface, allowing AI clients to discover and invoke AWS APIs with structured input/output validation without direct SDK knowledge.

Provides dedicated MCP servers for Terraform, AWS CDK, and CloudFormation that expose IaC operations as tools, enabling AI assistants to read, validate, plan, and apply infrastructure changes. The Terraform server parses HCL, the CDK server integrates with CDK CLI, and the CloudFormation server manages stack operations — each translating IaC-specific workflows into MCP tool schemas with structured input validation and change preview capabilities.

Manages MCP server startup, shutdown, and communication through stdio, SSE (Server-Sent Events), or custom transports. The MCP host (client) spawns server processes, establishes bidirectional communication channels, handles connection lifecycle (initialization, heartbeats, graceful shutdown), and manages resource cleanup. This enables reliable server operation with automatic restart on failure and clean shutdown semantics.

Provides an MCP server that exposes AWS documentation and developer guides as searchable resources, enabling AI assistants to reference official AWS documentation without external web searches. The server indexes AWS docs and enables semantic search over documentation content, allowing AI to provide accurate, up-to-date information about AWS services, APIs, and best practices.

Exposes database query execution and schema discovery as MCP tools through dedicated servers for PostgreSQL, DynamoDB, Neptune (graph), and Memcached. The PostgreSQL server uses SQLAlchemy for connection pooling and query execution with result streaming, DynamoDB server translates query patterns into DynamoDB API calls with scan/query optimization, and Neptune server handles Gremlin/SPARQL query execution — each providing structured schema introspection tools that allow AI assistants to understand data models before generating queries.

Exposes container management operations through dedicated MCP servers for ECS (task definition management, service scaling, container logs) and EKS (pod management, deployment operations, cluster introspection). The ECS server translates tool calls into ECS API operations with task lifecycle management, while the EKS server uses kubectl or Kubernetes Python client to manage workloads, enabling AI assistants to deploy, scale, and troubleshoot containerized applications without direct CLI knowledge.

Exposes AWS AI/ML services as MCP tools through dedicated servers: Bedrock server provides access to foundation models and knowledge base retrieval, SageMaker server enables notebook execution and model training/inference, Nova Canvas server handles image generation and editing. Each server translates tool calls into service-specific APIs with streaming support for long-running operations, allowing AI assistants to invoke other AI models, retrieve knowledge, and generate content without direct SDK calls.

Exposes AWS monitoring and operational data as MCP tools through dedicated servers for CloudWatch (metrics, logs, alarms), CloudTrail (audit logs), and Cost Explorer (cost analysis). CloudWatch server provides metric queries and log insights execution, CloudTrail server enables audit log filtering and analysis, and Cost Explorer server translates cost queries into structured API calls — allowing AI assistants to analyze operational health, security events, and spending without manual dashboard navigation.

Exposes IAM operations and policy analysis as MCP tools through a dedicated IAM server that can read existing policies, validate policy syntax, generate least-privilege policies, and simulate policy effects. The server uses IAM Policy Simulator API and policy document parsing to enable AI assistants to understand permission requirements, audit existing policies, and generate compliant IAM configurations without manual policy writing.

Exposes message queue and workflow operations as MCP tools through dedicated servers for SNS (publish/subscribe), SQS (queue management), and Step Functions (workflow execution and monitoring). SNS server handles topic management and message publishing, SQS server manages queue operations and message retrieval, and Step Functions server enables workflow execution and state machine introspection — allowing AI assistants to orchestrate asynchronous workflows and integrate with event-driven architectures.

Exposes object storage and search operations as MCP tools through dedicated servers for S3 (object operations, bucket management), S3 Tables (structured data queries), and Kendra (enterprise search). S3 server handles object upload/download/listing with metadata operations, S3 Tables server enables SQL-like queries on structured data, and Kendra server provides semantic search with document ranking — allowing AI assistants to access and search data without direct SDK knowledge.

Provides mechanisms for MCP clients to discover available servers, introspect their exposed tools and resources, and understand tool schemas and capabilities. The MCP protocol defines initialization handshakes and tool listing endpoints that allow clients to query server metadata, tool definitions (with input/output schemas), and resource types — enabling dynamic client adaptation to available server capabilities without hardcoded tool knowledge.