home / mcp / curl mcp mcp server

Curl MCP MCP Server

Provides a structured HTTP client for MCP with stdio and HTTP transports, including session handling, timeouts, and integration tests.

Installation

Add the following to your MCP client configuration file.

Configuration

View docs

{
  "mcpServers": {
    "calibress-curl-mcp": {
      "url": "http://localhost:3000/mcp",
      "headers": {
        "MCP_API_KEYS": "key1,key2",
        "MCP_REQUIRE_KEY": "true or false",
        "MCP_ALLOWED_HOSTS": "host:port,otherhost:port",
        "MCP_ALLOWED_ORIGINS": "https://yourapp.com"
      }
    }
  }
}

You can run curl-mcp to act as a dedicated HTTP/MCP client server that lets your AI assistants and MCP-aware tools send structured HTTP requests and receive rich, contextual responses. It supports both local stdio and hosted HTTP transports, complete with session handling, timeouts, redirects, and easy integration tests, so you can confidently build MCP-enabled workflows.

How to use

Set up curl-mcp and connect it to your MCP client. You can run a local stdio server for in-editor development or expose an HTTP endpoint for hosted clients. Your client will then issue requests via curl_request and receive structured responses that include status, timing, body, and optional advice.

How to install

Prerequisites: Node.js 20 or later is required for native fetch support.

From source, install dependencies and build the project.

npm install
npm run build

Start the stdio transport for local development and the HTTP transport for hosted clients.

npm run dev:stdio
npm run dev:http

Additional sections

Configuration and startup details are provided below to help you connect your MCP clients to curl-mcp in either stdio or HTTP mode.

Available tools

curl_request

A structured HTTP client designed for AI assistants and MCP-aware development tools. It handles HTTP verbs, headers, bodies, timing, redirects, and returns a rich, structured response with status, content, and optional advice.