sufetch

SuFetch provides an MCP server that lets AI assistants explore and generate code for a set of type-safe OpenAPI clients. The MCP server exposes a set of tools that enable discovering available APIs, inspecting endpoint details, and generating code samples in TypeScript that align with the underlying OpenAPI specifications. This is designed to help AI copilots understand your API surface quickly and produce accurate, typed requests and responses when integrating with services like Hetzner, DigitalOcean, Ory Kratos, and Ory Hydra. Once the MCP server is running, you can interact with its toolset to list APIs, fetch API metadata, search endpoints by path or method, retrieve schemas, and generate code examples tailored to your use cases.

Available tools include: list_apis (lists all APIs), get_api_info (fetches metadata for a given API), search_endpoints (filters endpoints by path/method/description), get_endpoint_details (provides full endpoint specs), get_schema_details (shows data schemas), generate_code_example (produces TypeScript code snippets), and get_quickstart (returns a quickstart guide for an API). These tools empower AI agents to reason about API capabilities and generate client code without manual browsing.

Prerequisites:

• Node.js and npm (or pnpm/yarn) installed on your system
• Git for cloning the repository (optional if using npm/global install)

Install globally (recommended for MCP usage):

# Install globally
npm install -g sufetch

# Verify installation
sufetch-mcp --version

If you prefer to build locally from source:

git clone https://github.com/productdevbook/sufetch.git
cd sufetch
pnpm install
pnpm build

Run the MCP server:

sufetch-mcp

This will start the SuFetch MCP server leveraging the prebuilt OpenAPI clients and MCP tools integrated into the package.

Note: If you are using a containerized environment, you can also run the MCP server via a Node-enabled image that has sufetch installed or built in the image build step.

Tips and common issues:

• Ensure Node.js version compatibility with the SuFetch package (refer to package.json engines in the repo).
• If the MCP server does not appear in Claude or your MCP UI, verify the command exposure (sufetch-mcp) and that the server is running on stdio transport as expected by your integration.
• When configuring Claude Desktop, you can set the MCP server with the command sufetch-mcp in CLAUDE_config.json, as demonstrated in the README.
• The MCP server supports multiple APIs through the sufetch package; if you add new OpenAPI specs to your project, rebuild to expose them via the MCP endpoints.
• Environment variables: typically not required for standard operation, but you may want to configure API keys or base URLs in your generated clients via the OpenAPI specs or client configuration as you integrate with AI workflows.