Agentic AI Engineer — Claude Agent SDK, OpenAI Agents SDK & MCP

I design and build AI agents that do real work inside your systems: they reason over your data, call your tools and APIs, follow guardrails, and hand off to a human when a decision needs one.

I'm Mudasser Tufail, a senior software consultant with 15+ years of experience. Today most of my work is agentic AI built on the Claude Agent SDK, the OpenAI Agents SDK and the Model Context Protocol (MCP), on top of the solid .NET, NodeJS and cloud engineering that production agents depend on.

What I do

AI agents that act, not just chat

A chatbot answers questions. An agent completes a task: it plans, calls tools, checks the result and decides what to do next. I build workflow-driven agents for engineering, operations and business processes, and I put most of the effort into the parts that make them dependable in production: clear tool contracts, structured outputs, deterministic guardrails, validation and human-in-the-loop approval.

A good example is the Incident Intelligence & Code Support Platform I work on. It picks up production incidents from Slack and Sentry, investigates the code with RAG-based project context and repository analysis, runs checks in ephemeral sandboxes, and proposes validated fixes as GitHub pull requests. When it can't resolve a case, it writes a structured investigation report instead of guessing.

Claude Agent SDK, OpenAI Agents SDK and MCP

I pick the agent framework to fit the problem rather than the other way round. The Claude Agent SDK suits agents that work with code, files and long-running tasks. The OpenAI Agents SDK is a good fit for multi-agent hand-offs and tight integration with OpenAI models. For explicit, stateful multi-step workflows I use LangChain and LangGraph. MCP lets any of them reach your systems through a standard interface.

  • MCP servers that expose your databases, internal APIs and documents to AI assistants and agents safely
  • Agents built with the Claude Agent SDK, the OpenAI Agents SDK and LangGraph, including tool calling and structured outputs
  • Retrieval-augmented generation (RAG) so agents answer from your own knowledge, with citations
  • Guardrails, validation and evaluation so you can measure behaviour before and after release
  • Integration with Slack, Sentry, GitHub and existing .NET, NodeJS or Python back ends

Technologies

The stack I use most for agentic AI work, alongside the back-end and cloud skills that keep agents running reliably:

  • Claude Agent SDK
  • OpenAI Agents SDK
  • Model Context Protocol (MCP)
  • Claude Code
  • LangChain
  • LangGraph
  • Python
  • LLMs
  • RAG
  • pgvector
  • Tool Calling
  • Structured Outputs
  • Guardrails
  • LLM Evaluation
  • NodeJS
  • NestJS
  • .NET Core
  • PostgreSQL
  • AWS
  • Docker

Relevant projects

Agentic and AI projects from my portfolio. Each case study covers the problem, the approach and the technology used.

How I work

  1. 1. Discovery

    We pick one workflow worth automating, agree what the agent may and may not do, and define how success will be measured.

  2. 2. Tools and data

    I map the systems the agent needs, expose them through MCP servers or typed tools, and set up retrieval over your documents.

  3. 3. Build with guardrails

    The agent is built with the right SDK, structured outputs, validation and approval steps, and tested against real examples.

  4. 4. Ship and improve

    We roll out to a small group, review traces and evaluation results, then widen access as confidence grows.

Frequently asked questions

What is an agentic AI engineer (or agentic AI developer)?

An agentic AI engineer builds AI systems that complete tasks rather than only answer questions. The agent plans steps, calls tools and APIs, checks its results and asks a human when it needs approval. Most of the work is engineering: tools, data access, guardrails, testing and deployment.

Should I use the Claude Agent SDK or the OpenAI Agents SDK?

It depends on the task and the models you want to use. The Claude Agent SDK is strong for agents that work with code, files and long-running tasks; the OpenAI Agents SDK is a good fit for multi-agent hand-offs on OpenAI models. I help you choose and can build with either, and MCP lets both reach the same tools.

Can you build an MCP server for our internal systems?

Yes. I build MCP servers that expose databases, internal APIs and documents to AI assistants and agents, with authentication, input validation and clear limits on what each tool can do.

How do you keep AI agents safe in production?

With structured outputs, deterministic guardrails, validation of every tool call, least-privilege access, human-in-the-loop approval for risky actions, and evaluation runs that show how the agent behaves before and after each change.

Do you work with international clients?

Yes. I work with clients and teams in different countries and industries, and my experience includes international work in China, Malaysia, US, Australia.

Have a workflow an AI agent could take over?

Tell me about the process, the systems involved and what success looks like. I'll reply with a practical first step.