vercel/next-devtools-mcp

Next.js Development for Coding Agent

What it solves

Coding agents (like Claude and Cursor) often lack real-time visibility into the state of a running application. next-devtools-mcp provides a bridge between these agents and a Next.js development server, allowing AI agents to see live runtime errors, logs, and route structures directly within their chat interface.

How it works

It acts as a thin connector using the Model Context Protocol (MCP). It automatically discovers running Next.js 16+ dev servers and proxies requests to the server's built-in MCP endpoint (/_next/mcp). It also provides "gateways" that direct agents to version-accurate documentation bundled within the project's node_modules or to the agent-browser CLI for browser automation.

Who it’s for

Developers using Next.js 16+ and AI-powered coding agents (such as Claude Code, Cursor, Gemini, or VS Code Copilot) who want their agents to have runtime context of their application.

Highlights

  • Live Runtime Access: Accesses current build, runtime, and type errors, as well as development server logs.
  • Automatic Discovery: Automatically scans common ports to find and connect to running Next.js dev servers.
  • Version-Accurate Docs: Points agents to the exact documentation matching the installed Next.js version found in node_modules.
  • Browser Automation Gateway: Facilitates the integration of the agent-browser CLI for automated browser testing.
  • Broad Agent Compatibility: Supports a wide range of MCP clients including Claude, Cursor, Codex, Gemini, and Warp.

Related

  • Project
    desplega-ai/agent-swarm

    agent‑swarm is an open‑source platform that turns a lead AI agent into a task‑routing hub, delegating work to containerised worker agents (Claude Code, Codex, etc.). It integrates with Slack, GitHub, email, and many other tools, persisting context via vector memory, and can automatically generate PRs, replies, or scheduled scripts. The system is model‑agnostic, runs via Docker (or Helm on Kubernetes), and is licensed under MIT.

  • Project
    cchenax/px-kvstore

    A lightweight, high-performance in-memory KV store optimized for deterministic LLM caching and fast local AI inference experimentation.

  • Project
    EvolvingLMMs-Lab/lmms-engine

    A unified, high-performance training engine for multimodal models that integrates distributed training, kernel fusion, and advanced optimizers to enable scalable pretraining and fine-tuning.

  • Project
    jonigl/mcp-client-for-ollama

    A terminal-based client that connects local Ollama LLMs to Model Context Protocol (MCP) servers, enabling local models to execute tools and access external resources.