APIFold is a tool that converts REST APIs into MCP (Model Context Protocol) servers, with the core goal of enabling AI agents to call existing APIs without additional coding. It reads OpenAPI 3.x or Swagger 2.x specifications and generates a real-time, production-ready MCP server endpoint that AI clients (such as Claude, Cursor, Copilot, or any MCP-compatible client) can connect to immediately. Tool calls execute real HTTP requests to real upstream APIs, credentials are stored securely, and the official emphasis is no stubs, no mocks, no glue code.
In terms of main capabilities, APIFold supports OpenAPI 3.x and Swagger 2.0, and can automatically and transparently convert Swagger 2.0 specifications. It provides a CLI tool, and via npx apifold serve you can start an MCP server locally with zero configuration. It has a built-in API registry containing 8 preconfigured API specifications: Stripe, GitHub, Slack, HubSpot, Twilio, OpenAI, Notion, and Petstore, supporting one-click deployment. For authentication, it supports OAuth 2.0, including Authorization Code (PKCE), Client Credentials, 8 provider presets, and automatic token refresh. Access control provides tool-level permission configuration, divided into three access profiles: read-only, read-write, and full access. It also provides analytics features, including call volume, latency percentiles, error classification, and usage quotas; supports custom domains (such as mcp.yourcompany.com) with DNS verification; each server has an encrypted and unguessable unique endpoint ID; supports both SSE and Streamable HTTP MCP transport modes; credentials are stored encrypted using AES-256-GCM; and it has SSRF protection, including DNS pinning, redirect blocking, and private IP rejection.
In terms of usage, running locally requires only one command to turn a specification into a running MCP server, using SSE transport, listening on localhost:3000, after which you can immediately connect Claude or Cursor, open http://localhost:3000 to access the dashboard. For the hosted platform, users can sign up at apifold.dev, import specifications (via URL, file upload, or browsing the registry), copy the connection snippet into Claude Desktop or Cursor, and the MCP server goes live and gets a unique endpoint URL. The core conversion logic is a standalone MIT-licensed npm package @apifold/transformer.
Applicable scenarios include: developers who want AI agents to directly call existing REST APIs; users who need to quickly connect third-party services (such as Stripe, GitHub, Slack, OpenAI, Notion) to MCP clients such as Claude and Cursor; and teams that need to run MCP servers locally or in a hosted manner.
Limitations and notes: official materials do not provide specific pricing, cost, or platform availability details, and none are currently provided. Code contributions must be made via pull request and merged after approval; security vulnerabilities should not be posted as public GitHub issues and should refer to the security policy. Regarding licensing, @apifold/transformer is MIT-licensed, and the rest is under the GNU Affero General Public License v3.0.
