MCP Fusion/Get started/Installation

Installation

Ask AI about Vinkius

Install @mcpfusion/core, scaffold your first server with the CLI and learn which add-on packages exist for auth, testing and deployment.

MCP Fusion ships as the npm package @mcpfusion/core with a set of optional add-ons. Everything runs on Node.js 18 or later with TypeScript 5.7 or later.

Install the core

bash

npm install @mcpfusion/core @modelcontextprotocol/sdk

The official @modelcontextprotocol/sdk is a peer dependency: MCP Fusion implements the MCP 2.0 protocol on top of it and stays fully compliant with it.

Scaffold a server

bash

npx @mcpfusion/core create my-server

The scaffold asks a few questions and generates a working project:

src/server.ts with initMCPFusion() ready to run
src/tools/ where file based routing discovers your tools
src/server.ts where startServer() picks the transport and port
A SKILL.md architectural contract that AI coding agents follow automatically

Work with an AI coding agent such as Cursor, Claude Code, Copilot or Windsurf? The CLI injects the SKILL.md contract into .cursorrules, .windsurfrules and .clinerules for you. The agent then writes MVA style code without being told. See Skills.

Run it locally

bash

mcpfusion dev --server ./src/server.ts

mcpfusion dev starts the server with hot reload. Connect a client to try it right away:

bash

claude mcp add my-server npx tsx src/server.ts

Add-on packages

Each add-on is a separate npm package under the @mcpfusion scope. Install only what you need.

Package	What it adds
`@mcpfusion/yaml`	Zero code tools written in YAML
`@mcpfusion/jwt`	JWT verification with HS256, RS256 and JWKS
`@mcpfusion/oauth`	OAuth 2.0 Device Authorization Grant
`@mcpfusion/api-key`	Timing safe API key validation
`@mcpfusion/skills`	Progressive `SKILL.md` disclosure
`@mcpfusion/testing`	In memory test harness for the whole pipeline
`@mcpfusion/vercel`	Deploy the same server to Vercel
`@mcpfusion/cloudflare`	Deploy the same server to Cloudflare Workers
`@mcpfusion/aws`	Turn Lambda functions and Step Functions into tools
`@mcpfusion/openapi-gen`	Generate Models, Views and Agents from OpenAPI specs
`@mcpfusion/prisma-gen`	Generate tools and Presenters from a Prisma schema

bash

npm install @mcpfusion/jwt

Server options

There is no separate config file: the server entry calls startServer() with a plain options object. The transport is stdio (default), http (Streamable HTTP with sessions, port, TTL, rate limits and CORS options) or stateless (MCP 2.0: one fresh server per request, no handshake, any load balancer can serve any call).

Next steps

Quickstart: build and connect a first tool
Deploy: put your connector on Vinkius Cloud for free
CLI: every mcpfusion command explained