OpenAPI to MCP Generator

OpenAPI is a standard format for describing REST APIs — it defines endpoints, parameters, request bodies, and responses. MCP (Model Context Protocol) is a standard for exposing tools to AI assistants. Converting OpenAPI to MCP means taking each API operation (GET /users, POST /orders, etc.) and turning it into an MCP tool definition — with a name, description, and inputSchema — so an AI assistant like Claude can call your API as a tool.

This generator supports OpenAPI 3.x (3.0.x and 3.1.x). It also accepts Swagger 2.x documents but the conversion is best-effort — some Swagger-specific fields may not map correctly. Both JSON and YAML input formats are accepted.

The operationId becomes the tool name — the stable, machine-readable identifier that MCP clients and models use to call your tool. Without an operationId, a name is generated from the HTTP method and path (e.g. get_users_user_id), which is functional but fragile — if your API paths change, the generated tool names change too, breaking any workflows that reference them by name. Always define operationIds in your OpenAPI spec.

$ref resolution across multiple files is not supported — all schemas must be inline in the document. anyOf/oneOf compositions include only the first variant and emit a warning. Header and cookie parameters are skipped (they are usually authentication-related, not model-facing). Authentication/security schemes are not converted. Response body schemas are not included in inputSchema (they appear only in the description).

Four things make a big difference: (1) Add operationId to every operation. (2) Write clear summaries — this becomes the tool description the model reads to decide when to call the tool. (3) Add descriptions to every parameter and request body field — models use these to know what value to pass. (4) Use enum on fixed-option fields (status, category, etc.) so the model picks valid values rather than guessing.

The TypeScript tab generates an MCP server file using the @modelcontextprotocol/sdk. Each tool gets a typed input interface and a server.tool() registration block with a TODO stub for the implementation. You can paste this directly into a new MCP server project and fill in the handler logic.

There are no server-side limits — conversion happens entirely in your browser. Very large specs (1000+ operations) may take a moment to process but there is no hard cap. Performance depends on your device.

No. The entire conversion runs in your browser using JavaScript. Your spec text is never sent to any server, logged, or stored anywhere. It exists only in your browser tab while the page is open.