@azure/mcp

Provides a TypeScript-based MCP server factory that handles protocol initialization, connection lifecycle, and graceful shutdown. Implements the Model Context Protocol specification with Azure-specific configuration patterns, managing server state transitions from startup through message handling to termination. Uses event-driven architecture to coordinate between transport layers and message handlers.

Provides a declarative schema system for defining tools and resources that MCP clients can discover and invoke. Uses JSON Schema for capability description with built-in validation to ensure tool definitions conform to MCP specification requirements. Supports typed input/output schemas with automatic validation before tool execution, preventing malformed requests from reaching handlers.

Implements MCP resource protocol allowing servers to expose files, documents, or context objects that LLM clients can read and reference. Uses a URI-based resource addressing scheme with MIME type support for different content formats. Clients discover available resources through the MCP protocol, enabling LLM context augmentation without embedding data directly in prompts.

Implements JSON-RPC 2.0 message routing with automatic request-response correlation and error handling. Routes incoming MCP messages to appropriate handlers based on method name, manages request IDs for async correlation, and provides structured error responses with detailed error codes and messages. Handles both synchronous and asynchronous handler execution with timeout management.

Provides pluggable transport layer supporting multiple communication protocols (stdio, HTTP, WebSocket) with automatic protocol negotiation. Abstracts underlying transport details from business logic, allowing servers to work across different deployment scenarios without code changes. Handles transport-specific concerns like framing, encoding, and connection management.

Implements MCP sampling protocol allowing servers to request LLM inference through connected clients. Enables servers to invoke LLM capabilities (text generation, reasoning) without maintaining separate LLM connections. Uses prompt templates with variable substitution and supports streaming responses for long-form generation.

Provides structured logging with automatic correlation IDs for tracing MCP requests end-to-end. Integrates with Azure Application Insights for metrics, traces, and error reporting. Logs all tool invocations, resource accesses, and protocol messages with configurable verbosity levels. Supports custom log sinks for integration with existing observability platforms.

Implements MCP protocol authentication with support for multiple credential types (API keys, OAuth2, managed identities). Enforces authorization policies at the tool and resource level, allowing fine-grained access control. Integrates with Azure AD for enterprise authentication and supports custom authorization handlers for domain-specific policies.

Supports invoking multiple tools in a single MCP request with automatic result aggregation and error isolation. Executes tools in parallel when possible, with configurable concurrency limits. Returns structured results with per-tool success/failure status, enabling clients to handle partial failures gracefully.

Provides configuration system supporting environment variables, configuration files, and programmatic setup. Validates configuration at startup and provides clear error messages for missing or invalid settings. Supports configuration inheritance and overrides for different deployment environments (dev, staging, production).