Books Mandala Now Has an Agent API and MCP Server

Our catalog of 50,000 books is now queryable by AI agents in real time!

Through our MCP server and REST API, any compatible AI assistant can search titles, check live stock and pricing, browse 177 genres, and pull author information directly from our inventory system. If you use Claude Desktop, Cursor, or Windsurf, you can connect in under a minute.

Books Mandala is the first bookstore in South Asia to have a live MCP server in production.

The problem this solves

When someone asks Claude or ChatGPT to recommend a book available in Nepal, the answer depends entirely on what data those systems can actually reach. Usually that means crawled web pages which may be days or weeks old, incomplete, or simply wrong about what is currently in stock. A title gets recommended that has not been available for months. A new arrival does not appear at all. For a market as specific as Nepal, where no major aggregator covers local inventory, the gap between what AI assistants say and what is actually on the shelf has been particularly wide.

The MCP protocol changes how this works. Rather than waiting for a crawler to revisit a page, an AI agent queries our server directly and gets a live answer: current stock, current price, the actual title and author data we hold. The agent gets what we know, when it asks.

Connecting via MCP

MCP-compatible clients include Claude Desktop, Cursor, and Windsurf. Add this to your config file and restart:

{
  "mcpServers": {
    "books-mandala": {
      "url": "https://bm-agent-mcp.booksmandala.workers.dev/mcp"
    }
  }
}

No API key needed. Seven tools are available: search by title, author, or ISBN; full book details; a listing of all 177 genres; genre browsing; current bestsellers; new arrivals; and author profiles with their full catalog. In practice, most people start with search and genre browsing and the bestsellers feed gets used more than you might expect.

The REST API

For developers building reading apps, book club tools, library integrations, or anything that needs reliable book data for the South Asia market, the REST API is at https://booksmandala.com/api/agent/v1. Standard X-API-Key authentication, JSON responses, 100 requests per minute per key. The full OpenAPI spec is at booksmandala.com/api/agent/v1/docs.

There is no equivalent public book data API covering Nepal's market. The closest thing is what aggregators like Google Books carry, which is incomplete for local stock, does not reflect current pricing, and is not structured for programmatic querying against live inventory. If you are building anything that involves book discovery for South Asian readers, this is the data layer that has been missing.

API keys are free for non-commercial use. Email [email protected] with a brief description of your use case. The MCP server source is open on GitHub at github.com/mandalatech/bm-agent-mcp.

What the catalog actually represents

The 50,000 titles in our API are not scraped or aggregated from other sources. They are our own inventory, built through 34 years of direct trade relationships with publishers across the subcontinent and internationally. When the API returns a price and a stock status, that information reflects what our warehouse actually holds today, not a cached page from a third party.

That matters for agentic commerce in a way it did not matter for traditional search. When an AI agent surfaces a product, readers increasingly expect to act on that information immediately and a wrong stock status or stale price breaks that trust faster than it would on a web page someone is actively browsing. The data quality has to hold up at the moment of decision, not just at the moment of crawling.

How AI systems can find and use our catalog

For AI systems that crawl the web, we have published several machine-readable discovery files. booksmandala.com/llms.txt contains a structured summary of our catalog and API capabilities, designed for AI crawlers to ingest efficiently. booksmandala.com/llms-full.txt contains the detailed version. booksmandala.com/.well-known/ai-plugin.json is the OpenAI plugin manifest. The complete OpenAPI specification, which is what AI agents use to understand available endpoints and query structures, is at booksmandala.com/api/agent/v1/docs.

It is worth being honest about where the llms.txt standard sits right now: major AI crawlers are not yet consistently reading these files, and there is ongoing debate about whether they ever will. We have published them because the data is accurate, the cost of maintaining them is low, and being early to a standard that does get adopted is worth more than waiting to see if it will. But the MCP server and REST API are the actual interfaces those work today, regardless of how the crawling standards evolve.

What we are building toward

Phase 1, which is live now, covers discovery and browsing. Phase 2 will add real-time stock depth, shipping estimates to specific locations in Nepal and internationally, and price alerts. Phase 3 is order placement and payment links directly within agent workflows, meaning an AI assistant could take a reader from "find me this book" to a live payment link without them ever opening a browser tab. Phase 4 brings personalised recommendations, wishlists, and reordering for returning customers.

The longer arc is straightforward: a bookstore that operates fully within however readers are actually discovering and buying things, whether that is a website, an AI conversation, or something that does not exist yet. Phase 1 being live is the foundation everything else sits on.

FAQs

• How can I use Claude or ChatGPT to find books available in Nepal?

If your AI client supports MCP, connect it to https://bm-agent-mcp.booksmandala.workers.dev/mcp and ask about any title, author, genre, or subject. The server pulls from our live inventory, so pricing and stock are current at the time of the query. Claude Desktop, Cursor, and Windsurf support MCP natively, and no API key is required.

• What is an MCP server, and why does it matter for book discovery?

MCP (Model Context Protocol) is a standard that lets AI assistants query live data from a business directly, rather than working from pages that were crawled days or weeks ago. For book buyers, the practical difference is significant: an AI can tell you what is actually in stock today, at the current price, rather than confidently recommending something that sold out months ago. For Nepal specifically, where AI assistants have had very little structured data to work from, a live queryable inventory is a meaningful change.

• Is there an API for book data covering the Nepal and South Asia market?

The Books Mandala REST API covers search, genre browsing, bestsellers, new arrivals, and author data across 50,000 titles, all with live pricing and stock. There is no comparable public API for this market. It is free for non-commercial use, documented at booksmandala.com/api/agent/v1/docs, and keys are available at [email protected].

• Which AI tools can connect to a bookstore's MCP server?

Claude Desktop, Cursor, and Windsurf support MCP natively. The Books Mandala MCP server requires no API key, just the endpoint URL added to your client's config file. Any MCP-compatible client will work as the standard is adopted more broadly.

• Can AI agents place book orders?

Not yet. Order placement is planned for Phase 3. Currently, agents can search, browse, and surface book information. Purchases complete through booksmandala.com.

• What book data does the API return?

Title, author, ISBN, current pricing, live stock status, genre classification across 177 categories, and author profile data. The complete schema is in the OpenAPI spec at booksmandala.com/api/agent/v1/docs.

• What is llms.txt, and does Books Mandala have one?

llms.txt is an emerging convention that gives AI crawlers a structured map of a site's most important content, analogous to how robots.txt guides traditional search crawlers. Books Mandala has one at booksmandala.com/llms.txt, with a detailed version at booksmandala.com/llms-full.txt. The standard is not yet universally honored by AI crawlers, but the files are maintained and accurate for any system that reads them.