Sami Ullah builds AI agents and Model Context Protocol (MCP) servers as part of full stack products. His work includes an MCP-controlled job application system, an AI voice calling agent for lead qualification, and LLM-generated crop advisories in Urdu, always with a human confirming the actions that matter.
I treat an AI agent as a normal part of the product: it gets typed tools, clear permissions and logs, and the risky steps (sending an application, starting a call) wait for a person to confirm.
MCP lets an assistant such as Claude call those tools directly. I have built MCP servers that expose a product's own actions, and I also wire LLMs from Anthropic, OpenAI and Google Gemini into NestJS and Next.js back ends where a chat window is not the right interface.
Official documentation: Model Context Protocol documentation, Anthropic API documentation
Own product
JobPilot is my own web app for remote job hunting. Every day it collects new remote jobs, scores each one against a master CV, writes a tailored CV and cover letter for good matches, and waits for a human to approve before anything is sent. It supports several job seekers in separate workspaces and can also be driven from an AI assistant through MCP.
Own product
A system I built that calls a list of consented contacts with an AI voice agent, holds a short qualifying conversation and records every result as a tracked outcome or lead. A NestJS API runs the dialing, compliance checks and webhooks, and a Next.js dashboard with a built-in AI assistant lets the owner run campaigns and ask for reports in plain language.
Own product
ReviewLens (working name) is a SaaS I built for online stores. It reads a store's customer reviews, returns and competitors and turns them into a short report: what customers complain about, which products cause the most problems and what to fix first. All numbers are computed in code, and AI only classifies text and explains the results.
The Model Context Protocol is an open standard that lets AI assistants such as Claude call tools exposed by your own software. An MCP server describes those tools with typed inputs, so the assistant can search data or trigger actions in your product without custom glue code for every assistant. Sami Ullah builds them for his own products.
Job-Pilot, a job application system he controls through an MCP server; an AI outbound calling and lead qualification system built on an ElevenLabs voice agent with a Gemini assistant in its dashboard; and, at Earth Scan Systems, a service that turns satellite data into crop advice written in Urdu by an LLM.
Agents only get the tools they need, every tool validates its input, every run is logged, and actions with real-world effects, such as sending an email or dialing a phone number, wait for an explicit human confirmation. Prompts and outputs are kept out of client-facing logs when they may contain personal data.