Indian quick-commerce company Zepto has unveiled an internal tool that lets users place Zepto Cafe orders using natural language commands via a large language model, such as Anthropic’s Claude. The system leverages MCP and Playwright to interpret instructions and execute orders seamlessly.
The tool, developed by a Zepto engineer and made available on GitHub, gained wider attention after co-founder and CEO Aadit Palicha highlighted it on LinkedIn.
The system uses MCP (Model Context Protocol), an open-source framework from Anthropic, which acts as a coordination layer between the AI model and live services, determining the appropriate actions and timing.
Unlike conventional chatbots that depend solely on pre-existing training data, MCP allows models to access up-to-date, permissioned information and execute specific workflows via a controlled interface. In Zepto’s implementation, MCP interprets user instructions and directs them to the corresponding action.
Playwright handles the execution of these actions, functioning as the browser automation layer.
Playwright operates a real web browser to navigate Zepto’s website, choose the delivery address, add items to the cart, and complete the order mimicking the actions of a human user instead of interacting directly with a backend API.
In a demo video, Pranav Chandra Prodduturi, senior category manager at Zepto, demonstrated Claude placing a Zepto Cafe dessert order to a selected address through the MCP server.
Some fascinating projects are coming to life in our office these days,” Palicha remarked, congratulating his colleague in a LinkedIn post.
The feature is not accessible through Zepto’s consumer app or website. To use the automation, users need to manually configure the MCP server and log in to Zepto via a web browser.
Regarding security, Zepto points out that phone numbers are transmitted through environment variables instead of being hardcoded. This ensures that sensitive identifiers never enter the codebase, version control, or logs.
Authentication is managed through a manual browser login, with sessions stored solely on the user’s local machine and never committed to the repository.
The system does not use any API keys or hard-coded credentials. Since Playwright runs within a real browser session, the automation leverages Zepto’s existing web security controls rather than bypassing them.
Similar experiments are emerging across India’s consumer internet ecosystem. For instance, Zomato recently developed an MCP server that allows users to place food orders via text-based prompts.
