Skip to content
UrbanChat Developers

AI agents

MCP server

The same catalogue, order and customer calls, as tools an AI agent can use directly over the Model Context Protocol. Nothing to wrap: connect, and the agent can answer “do you have this in blue, size L?” and place the order.

Connect

POSThttps://www.chatcom.in/api/v1/urbanchat/mcp

Transport
Streamable HTTP, stateless, JSON responses. Protocol 2025-06-18 and earlier.
Authentication
The company’s API key as Authorization: Bearer ucik_… (or X-API-KEY). Same company, rate limit and refusals as the REST API.
Sessions
None. GET and DELETE answer 405; every call is a POST.

Client configuration

For any MCP client that supports remote HTTP servers with headers:

mcp.json
{
  "mcpServers": {
    "urbanchat-ims": {
      "type": "http",
      "url": "https://www.chatcom.in/api/v1/urbanchat/mcp",
      "headers": {
        "Authorization": "Bearer ucik_xxxxxxxxxxxx_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
      }
    }
  }
}

Or from the command line, with Claude Code:

Terminal
claude mcp add --transport http urbanchat-ims https://www.chatcom.in/api/v1/urbanchat/mcp \
  --header "Authorization: Bearer ucik_xxxxxxxxxxxx_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"

To try the tools by hand, use the MCP Inspector and choose Streamable HTTP:

Terminal
npx @modelcontextprotocol/inspector

Tools

browse_catalogread
search, category, brand, in_stock, limit, offset. Products with options, variants, prices, stock and photos — as GET /catalog/products.
get_product_detailsread
id: a product id or SKU, or a variant sku or barcode. Every variant and photo — as GET /catalog/products/{id}.
list_categoriesread
Categories and brands with product counts.
find_customerread
phone, recent_orders. As GET /customers.
get_orderread
reference, phone. As GET /orders/{reference}; pass the phone whenever answering a customer.
place_orderwrite
The POST /orders body. Creates a confirmed order and reserves stock. Idempotent on external_reference. The only tool that writes.
search_products, get_productread
The v1 flat search, kept for agents built against it. Prefer browse_catalog.

Results are JSON text content, in exactly the REST response shapes. A tool that cannot answer returns isError: true with the REST error object, e.g. { "error": { "code": "insufficient_stock", … } }.

Prompting an agent well

  • Ask the customer for each option the product lists in options, then order the variant whose attributes match. Order by the variant’s sku.
  • Check the variant’s available before promising anything. If an order comes back insufficient_stock, offer what is available.
  • Collect a delivery address with its PIN code and whether the customer is paying cash on delivery, and pass them as delivery_address and payment, so the parcel ships without anyone retyping it.
  • Give every order an external_reference derived from the conversation, and reuse it on a retry.
  • Read the order back from place_order’s result — it includes the lines, total and address — before confirming to the customer.
  • Pass the customer’s phone to get_order, always.