mxcp

Generates complete Model Context Protocol (MCP) server implementations from declarative YAML configuration files, eliminating boilerplate code generation. The framework parses YAML schemas defining tools, resources, and prompts, then auto-generates Python server code with proper MCP protocol compliance, type validation, and error handling built-in. This approach reduces MCP server development from hundreds of lines of manual code to configuration-only definitions.

Automatically converts SQL queries into callable MCP tools with intelligent parameter extraction, type inference, and result formatting. The framework parses SQL statements to identify input parameters (via placeholders or named parameters), infers types from database schema, and generates tool schemas with proper input validation and output serialization. This enables exposing arbitrary SQL queries as LLM-callable functions without manual schema definition.

Implements declarative access control policies that are evaluated at the MCP server level before tool execution, supporting role-based access control (RBAC), attribute-based access control (ABAC), and policy-as-code patterns. Policies are defined in YAML or Python and integrated into the request pipeline, allowing fine-grained control over which users/clients can invoke which tools or access which data. Authentication integrates with standard providers (OAuth2, API keys, JWT) and custom backends.

Enables defining data transformation pipelines using YAML or Python DSL, supporting multi-step workflows with SQL transformations, Python functions, and data validation. Pipelines can be triggered on schedules, events, or manual invocation, with built-in support for error handling, retries, and state management. The framework orchestrates pipeline execution, manages intermediate data, and provides observability into pipeline runs.

Provides structured logging, metrics collection, and tracing for all MCP server operations including tool invocations, authentication events, and pipeline executions. Logs are emitted in structured JSON format with configurable sinks (stdout, files, external services), and metrics can be exported to monitoring systems. Tracing captures request flow through the server with timing information, enabling performance analysis and debugging.

Automatically generates MCP-compliant tool schemas from Python type hints, SQL parameter types, or YAML definitions, with runtime validation of tool inputs and outputs. The framework uses Python's typing module and database introspection to infer parameter types, generate JSON Schema representations, and validate incoming tool calls against the schema before execution. This ensures type safety across the LLM-to-tool boundary.

Implements MCP server protocol compatible with multiple LLM clients (Claude, ChatGPT, local models via Ollama, etc.), abstracting away client-specific protocol variations. The framework handles protocol negotiation, capability advertisement, and response formatting for different clients, allowing a single MCP server to serve multiple LLM platforms without client-specific code.

Provides a framework for defining and managing reusable MCP resources (documents, templates, data) and prompt templates that can be referenced by tools or LLM clients. Resources are versioned, can be updated without server restart, and support dynamic content generation. Prompt templates support variable interpolation and can be composed to build complex prompts for LLM execution.

Implements comprehensive error handling throughout the MCP server with structured error responses, retry logic, and fallback mechanisms. Tool execution errors are caught and formatted as MCP-compliant error messages, authentication failures are logged and reported, and pipeline failures trigger configurable recovery actions. The framework supports exponential backoff retries for transient failures and circuit breaker patterns for cascading failures.

Supports environment-specific configuration (development, staging, production) with support for environment variables, configuration files, and secrets management. The framework loads configuration at startup, validates it against a schema, and provides runtime access to configuration values. Secrets are handled securely with support for external secret stores (AWS Secrets Manager, HashiCorp Vault, etc.).