@orval/mcp

Generates fully-typed TypeScript client code from OpenAPI 3.0+ specifications, using Model Context Protocol (MCP) as the transport layer for LLM-driven code generation workflows. Parses OpenAPI schemas into an intermediate AST representation, then templates TypeScript with proper type inference for request/response payloads, query parameters, and path variables. Integrates with Claude and other MCP-compatible LLMs to enable AI-assisted API client generation and modification.

Registers OpenAPI operations as callable MCP tools with full JSON Schema definitions for inputs and outputs, enabling LLMs to discover and invoke API endpoints through the MCP tool-calling interface. Converts OpenAPI parameter definitions (path, query, body, header) into MCP input schemas with proper validation constraints (required fields, type constraints, enum values). Handles request/response serialization and error mapping back to the LLM.

Maintains type consistency between OpenAPI schemas and generated TypeScript types through a two-way mapping system. Parses OpenAPI definitions into an intermediate representation, generates TypeScript interfaces/types with proper nullability and optionality inference, and can reverse-engineer TypeScript types back into OpenAPI schema updates. Detects schema drift and provides migration guidance when APIs change.

Provides configuration and lifecycle management for running @orval/mcp as an MCP server, handling server initialization, tool registration, request routing, and graceful shutdown. Supports both stdio and HTTP transports for MCP communication, manages environment variables and API credentials, and provides logging/debugging hooks. Integrates with Claude Desktop and other MCP clients through standard MCP server discovery mechanisms.

Implements middleware pipeline for transforming API requests (parameter serialization, header injection, auth) and responses (deserialization, error mapping, retry logic) before passing to LLMs. Supports custom transformers for request/response mutation, automatic error classification and retry strategies (exponential backoff, circuit breaker), and response normalization to ensure consistent LLM-consumable output. Handles HTTP status codes, timeout errors, and API-specific error formats.

Validates incoming LLM tool calls against OpenAPI schema constraints (required fields, type validation, enum values, min/max bounds, pattern matching) before executing API requests. Uses JSON Schema validation with OpenAPI-specific extensions (discriminators, oneOf/anyOf resolution, format validation). Provides detailed validation error messages to LLMs for constraint violations, enabling LLMs to self-correct malformed requests.

Manages multiple OpenAPI specifications and API integrations within a single MCP server, enabling LLMs to compose tool calls across different APIs. Provides namespace isolation for tools from different APIs, handles cross-API dependencies (e.g., using output from API A as input to API B), and manages shared state/context across API calls. Supports tool grouping and discovery filtering to reduce cognitive load on LLMs.

Supports hot-reloading of OpenAPI specifications without restarting the MCP server, enabling dynamic updates to available tools as APIs evolve. Tracks OpenAPI spec versions, detects breaking changes (removed operations, type changes), and provides migration guidance. Allows LLMs to query available API versions and choose which version to use for tool calls, supporting gradual API deprecation.

Generates concise, LLM-optimized documentation for API tools from OpenAPI specs, including operation summaries, parameter descriptions, example requests/responses, and usage patterns. Provides tool discovery and filtering mechanisms (by tag, operation type, required authentication) to help LLMs find relevant tools. Includes example-based learning where LLMs can see successful API call patterns before attempting their own.