vercel-labs/mcp-handler

Easily spin up an MCP Server on Next.js, Nuxt, Svelte, and more

What it solves

mcp-handler is a framework-agnostic HTTP adapter that allows developers to host Model Context Protocol (MCP) servers in JavaScript and TypeScript applications. It removes the need to write framework-specific boilerplate when deploying MCP servers to modern web frameworks that use Web-standard Request and Response APIs.

How it works

The library provides a createMcpHandler function that converts an MCP server definition into a standard Web-standard (Request) => Promise<Response> handler. This handler can then be mounted directly into routes in frameworks like Next.js, Nuxt/Nitro, SvelteKit, and Hono. It natively supports the 2026-07-28 MCP specification (stateless) and provides a transparent fallback for 2025-era Streamable HTTP clients.

Who it’s for

Developers building MCP servers using JavaScript or TypeScript who want to deploy them on serverless or edge-compatible frameworks that support Fetch-compatible APIs.

Highlights

  • Framework Agnostic: Works with any framework that exposes Web-standard Request and Response APIs.
  • Dual-Protocol Support: Handles both the latest 2026-07-28 stateless protocol and 2025-era Streamable HTTP from a single handler.
  • Simplified Deployment: Eliminates the need for Redis or complex session management for stateless operations.
  • TypeScript Support: Includes full type definitions for a type-safe development experience.

Related

  • Project
    thu-nics/C2C

    Cache‑to‑Cache (C2C) is a research‑grade Python library that lets two large language models exchange information by translating their KV‑cache tensors instead of generating text. Pre‑trained projector “fusers” for several Qwen/Llama model pairs are provided, and the framework includes scripts for inference, interactive chat, training new projectors, and evaluation. It claims 8‑10 % higher accuracy and ~2× lower latency compared to text‑based model‑to‑model communication.

  • Project
    lablup/backend.ai

    A container-based computing cluster platform that orchestrates ML frameworks and provides multi-tenant access to heterogeneous hardware accelerators like GPUs and NPUs.

  • Project
    Acly/comfyui-tooling-nodes

    A set of ComfyUI custom nodes and API extensions that enable the use of ComfyUI as a backend for external tools, featuring optimized image transfer, regional prompting, and model inspection.

  • Project
    litexlang/golitex

    Litex is a set-theory-based formal language designed for readable, machine-checked mathematics that allows users to write facts in a natural order and optionally compiles to Lean.