Feed in → agent-ready storefront out

Your product feed, ready for AI shopping agents.

AgentsFront turns a Google Shopping feed into your own MCP endpoint — so assistants like Claude can search your catalog and answer real questions about your products, on an endpoint you control. No plugin, no platform lock-in.

Try the live demo → View the source
A functional experiment and public proof point — open source, Apache-2.0.
Live demo

Shop a real store, from Claude.

Clay to Clay is a real ceramics shop with a real product feed. Add its endpoint to Claude and shop it in plain language — you'll get actual products, prices, and images back.

1

Add the connector

In Claude → Settings → Connectors → Add custom connector, paste the endpoint URL.

https://mcp.claytoclay.com/mcp
2

Ask like a shopper

Start a chat and ask naturally — for example:

"handmade cups, a pair, under $100"
3

Get real results

Claude calls the catalog tools and answers with matching products, prices, and image links — straight from the feed.

How it works

One catalog, two lanes.

A scheduled lane keeps your catalog fresh from the feed. A serving lane answers agents on demand. The catalog in the middle is the only shared piece.

Source
Your store
Google Shopping XML feed
→
Ingest
Catalog
Postgres + search + embeddings
→
Serve
MCP gateway
search · lookup · get_product
→
Consume
Claude / agent
answers the shopper
Search that understands intent. Full-text out of the box, plus optional semantic search (pgvector embeddings) so "something for baking bread" finds the right thing even when the words don't match.
Self-host

Run it yourself in ~10 minutes.

Have a Google Shopping feed and Docker? Point it at your feed and you get a working MCP endpoint locally — no cloud account required. 

# clone, then point it at your feed
git clone https://github.com/asharmairaj/agentsfront
cd agentsfront

AGENTSFRONT_FEED_URL="https://yourstore.example/feed.xml" \
  docker compose up --build

# → MCP endpoint at http://localhost:8080/mcp
  • Zero cloud lock-in. Postgres + pgvector in Docker; full-text search works immediately.
  • Pluggable embeddings. Off by default, or point at any OpenAI-compatible endpoint (e.g. a local Ollama).
  • Production path included. The same code deploys to Google Cloud Run with Cloud SQL + Vertex.
  • Apache-2.0. Fork it, host it, make it yours.