What is AI-enabled web development?

It's the new layer of full-stack work in 2026 — large language models, agents, and ML inference wired into both the product (chat, copilot, summarization, semantic search) and the engineering loop (Claude Code plugins, MCP servers, local-first coding agents). I've been shipping it for the last year or two across Next.js, Node.js, and the MERN stack. The high-leverage piece is almost always the typed integration surface, not the prompt.

What is an MCP server?

An MCP (Model Context Protocol) server is just a process that exposes tools, data, and capabilities to an LLM client over a standardized protocol — kind of like a GraphQL endpoint, but for an agent. It lets a client like Claude Code read your database, call your APIs, or run domain-specific actions without bespoke integration code. I build custom MCP servers, including the open-source MERN MCP, that automate full-stack CRUD scaffolding and developer workflows.

What is a Claude Code plugin?

A Claude Code plugin extends the Claude Code CLI with custom commands, hooks, MCP servers, sub-agents, or skills tailored to your codebase. Plugins can narrate edits, enforce conventions, run pre-commit checks, or wire in proprietary tools. I've authored claude-human-review, which describes each edit Claude is about to make in plain English so engineers can approve or undo with full context — diff hunts are the wrong review surface.

What is a ReAct agent?

A ReAct (Reasoning + Acting) agent is an LLM that interleaves thoughts with concrete tool calls — read a file, run a search, write a patch — until a task is complete. It's the loop behind most modern coding agents. My open-source project nip is a local-first ReAct agent powered by Gemma on Ollama, with no cloud calls and no vendor lock-in.

Is Irteza Asad Rizvi available for hire?

Yes. I'm open to senior full-stack roles, AI-focused contracts, and consulting engagements. Remote out of Toronto, specializing in shipping production AI features — LLM agents, Claude Code plugins, MCP servers — into Next.js, Node.js, Vue, and Nuxt apps. Fastest way to reach me: irtezaasad@gmail.com.

What stack does Irteza work in?

Primary: TypeScript, Node.js, Next.js, React, Vue, Nuxt, Python, MongoDB, PostgreSQL, Docker, AWS, Linux. AI-specific: Claude SDK, Model Context Protocol, Ollama, ONNX Runtime. CI/CD: Jenkins and Docker. Seven-plus years of production web work and active AI tooling on top.

MCP server · Open source · Full-stack scaffolding

MERN MCP — Full-Stack CRUD Scaffolding for AI Agents

MERN MCP is a Model Context Protocol server I wrote so my coding agents could do the boring half of full-stack CRUD without me typing it. Point an MCP-aware client at the server, ask it to add an Order resource, and you get the Mongoose schema, the Express route, the React form, and a preview diff to approve before any file actually changes. The agent can't drift on field names, because the server is authoritative about what the field names are.

[TypeScript][Node.js][Express][React]

View on GitHub Hire me for similar work

Why I built it

Eighty percent of MERN feature work is the same shape: model, route, validator, form, list view, detail view. Hand-rolling that shape every time is a tax on real product thinking. Modern code agents will happily write it for you, but they tend to drift — slightly wrong field names, slightly wrong Mongoose conventions, slightly wrong test setup, the kind of subtle wrongness that takes longer to fix than to write from scratch.

An MCP server fixes the drift by making the server authoritative. The schema lives in one place and the agent's tool calls are typed against it.

How it works

The server exposes tools like generate_model, generate_route, generate_react_form, and apply_patch. Each tool returns a structured proposal — files, hunks, descriptions — and the agent must call apply_patch explicitly to commit. The preview-before-apply boundary is the leash.

It plugs into any MCP-compatible client (Claude Code, custom agents, your own LangChain wrapper) and ships with project templates so you can boot a new MERN app, scaffold five resources, and have a running CRUD UI in under ten minutes.

What I take from it

MCP servers are, I think, the right shape for AI-enabled web development. Instead of giving the agent free run of a codebase, you give it a typed surface that knows your conventions. As far as I can tell, MCP server development is the highest-leverage skill a senior full-stack engineer can pick up right now — every team has 5–10 internal procedures that would be better as typed agent tools than as tribal knowledge.

← All projects More on AI engineering →Writing on AI & engineering →