MCP catalog search: ChatGPT/Claude shopping your index
Use Seekmodo MCP so ChatGPT and Claude can search your catalog with the same tenant-safe index as your storefront.
Shoppers are not the only clients of your catalog anymore. Support copilots, internal ops agents, and consumer AI assistants increasingly expect to query product data — not scrape HTML. Model Context Protocol (MCP) is how Seekmodo exposes catalog search to those agents with the same tenant isolation you already use for the storefront.
This playbook explains what MCP catalog search is for, how authentication works at a high level, and when to enable it.
The problem MCP solves
Without a protocol, teams bolt ChatGPT or Claude onto a store by pasting CSV exports, building one-off scrapers, or granting broad database access. Those approaches go stale, leak data, or invent products that are not for sale.
MCP lets an agent call tools such as search against your Seekmodo index — the same relevance stack typeahead and SERP use — instead of improvising. The agent gets structured results grounded in what you indexed.
Storefront discovery vs operator MCP
Seekmodo separates concerns:
- Anonymous / storefront MCP discovery — well-known endpoints and meta tags so public agents can find a tenant’s MCP URL for a tight allowlist (for example search and handshake), with rate limits. See the connectors public discovery notes on the docs site.
- Authenticated operator MCP — bearer tokens minted from admin for Claude Desktop, Cursor, and similar clients, with human approval patterns for write tools.
HMAC-authenticated connector traffic and operator tokens stay on the gateway paths documented in /docs/mcp. Do not paste secrets into blog posts or tickets.
Who should care
- Merchants who want ChatGPT/Claude to answer “do you have X in stock?” from live index data
- Agencies building internal copilots over multiple client catalogs
- Teams standardizing on MCP across CMS connectors (Woo, Zen Cart, Magento, BigCommerce, WordPress, MediaWiki)
MCP is included on Seekmodo paid plans from Hobby up — a deliberate contrast with vendors that do not publish an MCP server at any tier (see footnoted claims on /compare/doofinder).
How it feels day to day
An operator mints a token in admin, pastes a client snippet into Cursor or Claude Desktop, and asks “list my Seekmodo tools” or runs a playbook like fix-zero-results. Write actions still expect human confirmation in the product flow. Shopper-facing search on the site does not require the merchant to run an MCP client.
Security posture (plain language)
Tenant isolation is the product. Agents should not see another merchant’s catalog. Prefer least-privilege tokens, rotate when laptops change hands, and keep storefront anonymous MCP on the narrow tool allowlist. Full detail lives in docs — this post is the why, not the runbook.
What good looks like in the first hour
After minting a token, you should see Seekmodo tools listed in your MCP client, successfully call a read-only search against a known SKU or page title, and confirm the result matches the storefront SERP closely enough that you trust the agent. If the agent invents products, you are not grounded in the index — fix auth, tenant, or tool selection before you demo to a customer.
Write tools (synonyms, pins, playbooks) should still feel gated by human approval. That friction is a feature for production catalogs.
Next step
Read /solutions/mcp-catalog-search, the implementer guide at /docs/mcp, and the operator mint runbook linked from admin. Start a trial from /signup if you want to wire an agent against a sandbox tenant first.