Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Zepto MCP Server order 2 Amul milk and 1 brown bread to my home address, pay with COD"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
š Project Overview
The project is a verified implementation of Anthropic's Model Context Protocol integrating Claude Desktop with Zepto's live e-commerce platform. The workflow completed a live Cash on Delivery order of ā¹79 with zero human intervention required for store routing errors.
Related MCP server: RappiMCP
Traditional grocery ordering requires navigating numerous user interface screens: setting addresses, searching products, comparing weights, managing carts, selecting payment methods, and reviewing fees.
This project explores Conversational Commerce: enabling an LLM (Claude) to execute the complete grocery ordering journey while:
Grounding Intent: Translating loose natural language requests (e.g., "order me dairy milk roast almond chocolate price 49") into verified product SKUs using historical order frequency.
Autonomous Error Recovery: Detecting API dependency failures (such as "Store not selected") and self-healing by resolving saved addresses to bind the dark store.
Strict Human Safeguards: Implementing a non-destructive preview phase (confirm: false) followed by explicit human confirmation (confirm: true) to ensure zero unauthorized financial transactions occur.
The integration bridges Claude Desktop with Zepto's live quick-commerce microservices using the open Model Context Protocol (MCP) standard.
flowchart TD
User([š¤ User / Natural Language Prompt]) <--> Claude[š§ Anthropic Claude Desktop]
Claude <-->|JSON-RPC via stdio| Bridge[š mcp-remote Bridge]
Bridge <-->|Encrypted HTTPS / SSE Stream| ZeptoServer[āļø Zepto MCP Server]
ZeptoServer <-->|Internal REST APIs| ZeptoCore[(š¢ Zepto Quick-Commerce Core<br/>- Dark Store Routing<br/>- Live Catalog & Inventory<br/>- Cart & Checkout Engine)]
classDef primary fill:#e1f5fe,stroke:#0288d1,stroke-width:2px,color:#01579b;
classDef secondary fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px,color:#4a148c;
classDef server fill:#e8f5e9,stroke:#388e3c,stroke-width:2px,color:#1b5e20;
class User,Claude primary;
class Bridge secondary;
class ZeptoServer,ZeptoCore server;The integration utilizes 8 production tools exposed by the Zepto MCP server:
Tool Name | Parameters | Purpose |
| None | Fetches recent orders with delivery status ( |
| None | Returns unique past products ranked by purchase frequency to disambiguate loose queries. |
| None | Retrieves user-saved delivery locations (e.g., "Home"). |
|
| Binds the active session to a saved address, which automatically locks the serving dark store. |
|
| Searches live inventory in the selected dark store. |
|
| Adds, modifies, or removes cart items. |
| None | Returns available payment channels (COD, Web Link, Wallet) and total charges. |
|
| Two-phase checkout execution (Preview vs Final Placement). |
sequenceDiagram
autonumber
actor User as š¤ User
participant Claude as š§ Claude Desktop
participant Zepto as āļø Zepto MCP Server
User->>Claude: "Order me dairy milk roast almond chocolate 49"
Claude->>Zepto: get_past_order_items()
Zepto-->>Claude: Ranked list (Dairy Milk Almond top item - 13 orders)
Claude->>Zepto: search_products("Cadbury Dairy Milk Roast Almond")
Note over Zepto,Claude: ā ļø ERROR: "Store not selected"
rect rgb(255, 245, 230)
Note over Claude: [AUTONOMOUS SELF-HEALING RECOVERY]
Claude->>Zepto: list_saved_addresses()
Zepto-->>Claude: Found saved address "Home"
Claude->>Zepto: select_saved_address(addressId)
Zepto-->>Claude: Store context locked to nearest dark store
end
Claude->>Zepto: search_products("Cadbury Dairy Milk Roast Almond") [RETRY]
Zepto-->>Claude: 10 items returned -> Matched 36g SKU @ ā¹49
Claude->>Zepto: update_cart(quantity: 1, price: 4900 paise)
Zepto-->>Claude: Cart updated successfully
Claude->>Zepto: get_payment_methods()
Zepto-->>Claude: Total: ā¹79 (Item ā¹49 + Delivery Fee ā¹30)
Claude->>User: "Total is ā¹79. Pay via Cash on Delivery?"
User->>Claude: "Yes, proceed with COD"
Claude->>Zepto: create_order(confirm: false) [PREVIEW]
Zepto-->>Claude: Preview verified
Claude->>Zepto: create_order(confirm: true) [COMMIT]
Zepto-->>Claude: Order Placed! Status: INITIATED
Claude->>User: ā
Order successfully initiated!zepto-mcp-assistant/
āāā .gitignore # Standard ignore patterns for Node, Windows & secrets
āāā LICENSE # MIT Open-Source License
āāā package.json # Project metadata and diagnostic scripts
āāā README.md # Main repository documentation
āāā config/
ā āāā claude_desktop_config.example.json # Sanitized Claude Desktop configuration template
āāā docs/
ā āāā assets/
ā ā āāā live_order_tracking.png # Live order tracking proof from Zepto web app
ā ā āāā delivered_product_proof.jpg # Physical delivered product photo
ā āāā ARCHITECTURE.md # Deep-dive system architecture & data contracts
ā āāā DEMO_SCRIPT.md # Concise 2ā3 minute presentation script & proof links
ā āāā PRESENTATION_DECK_NOTES.md # Presentation deck notes & Google Gemini link
āāā scripts/
āāā test_connection.bat # One-click diagnostic script to verify server connectionOperating System: Windows 10/11, macOS, or Linux.
Node.js: v18.0.0 or higher (node -v $\ge$ 18.0; verified on Node v24.21.0).
AI Client: Claude Desktop installed.
Zepto Account: An active Zepto account with at least one saved delivery address in the mobile app.
Open your Claude Desktop configuration file:
Windows:
%APPDATA%\Claude\claude_desktop_config.json(Or if installed via Windows Package Manager / MSIX):
%LOCALAPPDATA%\Packages\Claude_pzs8sxrjxfjjc\LocalCache\Roaming\Claude\claude_desktop_config.jsonmacOS:
~/Library/Application Support/Claude/claude_desktop_config.jsonAdd the following entry under "mcpServers":
{
"mcpServers": {
"zepto": {
"command": "cmd",
"args": [
"/c",
"npx",
"-y",
"mcp-remote",
"https://mcp.zepto.co.in/mcp"
]
}
}
}You can test the connection directly by double-clicking scripts/test_connection.bat or by running:
npx -y mcp-remote https://mcp.zepto.co.in/mcpCompletely restart Claude Desktop. The hammer icon in the bottom-right corner will show the zepto server with 8 available tools.
š Live Working Demonstration & Verification
This project successfully executed an end-to-end, live grocery purchase on Zepto. The order was placed conversatially via Claude Desktop, automatically recovered from a store-selection error, dispatched to a live delivery rider, and delivered directly to the doorstep.
1. Live Order Tracking (Zepto Web) | 2. Physical Delivered Product (Doorstep Proof) |
|
|
Order ID: | Item: Cadbury Dairy Milk Roast Almond (36g)Total Paid: ā¹79 (Cash on Delivery) |
Test Scenario | Action Tested | Observed Result |
History Retrieval | Fetch past user orders | Passed: 8 orders parsed and converted from paise to INR. |
Out-of-Scope Request | Request to rate order | Passed: Handled gracefully; directed user to official app. |
Store Dependency | Search catalog without location | Passed: Caught |
SKU Disambiguation | Match loose query with price | Passed: Correctly matched ā¹49 variant (36g) against premium variants. |
Transaction Safeguard | Execution without consent | Passed: Enforced preview and halted for human confirmation before spending. |
Non-Destructive Previews: Transaction endpoints require a two-stage commit (confirm: false followed by confirm: true).
Human-in-the-Loop Safeguards: The model halts before checkout to disclose item costs and delivery fees.
Data Scrubbing: No personal phone numbers, physical addresses, or active session tokens are committed to this repository.
š System Architecture & Tool Contracts: Deep dive into JSON-RPC messages and schema definitions.
šļø Concise Demo Presentation Script: 2ā3 minute presentation walkthrough with live proof links.
š„ Live Video Recording (Google Photos): Screen recording and delivery proof in Google Photos.
š Interactive Presentation Deck (Gemini): View the interactive slide deck on Google Gemini.
This project is open-source and available under the MIT License.
This server cannot be deployed
Agentic commerce with 58 MCP tools for product search, checkout, A2A negotiation, C-Suite analytics.
Run your ecommerce ads from Claude & ChatGPT: Meta, Google, Amazon, Shopify (150 tools)
Directory of APIs, merchants, and tools AI agents can actually use.
Bounded tools for rendering, extraction, RAG, enrichment, local discovery and review analysis.
