@elixa/mcp-server

The universal shopping MCP for AI agents. One install, search across every connected merchant backend — ranked by quality, completeness, and trust.

• No merchant fees. Merchants are free to be discovered.
• No user fees. The MCP is free to install and use.
• Neutral ranking. Results are ordered by a composite of relevance, data completeness, merchant trust (Tranco domain authority), price-fit, and rating — not by who paid most.
• Runs over stdio. Works with Claude Desktop, Cursor, Zed, and every MCP-compatible client.

Install

Claude Desktop

Edit claude_desktop_config.json:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the Elixa entry:

{
  "mcpServers": {
    "elixa": {
      "command": "npx",
      "args": ["-y", "@elixa/mcp-server"]
    }
  }
}

Restart Claude Desktop. Ask it "find me a wool jumper under £80" and watch it call the search_products tool.

Cursor / Zed / other MCP clients

Same pattern — add an elixa entry under mcpServers pointing at npx -y @elixa/mcp-server.

Tools exposed

search_products

Search every connected backend in parallel. Returns ranked, deduplicated results.

Arguments:

| Arg | Type | Notes | |---|---|---| | query | string | Natural-language shopping query (required) | | min_price, max_price | number | Numeric price band | | currency | string | ISO 4217 code (USD, GBP, EUR) | | brand | string | Brand-name filter | | category | string | Category filter | | merchant | string | Merchant domain filter | | availability | "in_stock" / "any" | Restrict to in-stock results | | min_trust | 0–100 | Minimum merchant trust score | | min_completeness | 0–100 | Minimum product data completeness | | min_rating | 0–5 | Minimum product rating | | sort | relevance / price_asc / price_desc / rating / trust | Sort order | | limit | 1–50 | Default 20 |

get_product

Get the full product record for an elixa_id returned from search_products.

find_alternatives

Given a product, return similar alternatives — cheaper by default, or from higher-trust merchants.

Backends

| Backend | Status | Coverage | |---|---|---| | Elixa (our index) | Always on | WooCommerce, Magento, custom storefronts, DTC opt-ins | | Shopify Catalog MCP | Opt-in via env vars | All eligible Shopify stores | | Amazon PAAPI | Roadmap | Amazon US catalog | | Etsy Open API | Roadmap | Handmade + vintage | | eBay Finding API | Roadmap | Resale + refurbished |

Configuration

Everything's env-var driven. Defaults point at the live Elixa API so the server works out of the box.

| Env var | Default | Notes | |---|---|---| | ELIXA_API_BASE | https://api.elixa.app | Root of the Elixa REST API | | ELIXA_API_KEY | none | Optional; Elixa's free tier is unauthenticated | | SHOPIFY_CATALOG_MCP_URL | none | Shopify's Catalog MCP endpoint — enables the Shopify backend | | SHOPIFY_CLIENT_ID | none | Shopify Dev Dashboard → Catalog section | | SHOPIFY_CLIENT_SECRET | none | ditto | | BACKEND_TIMEOUT_MS | 5000 | Per-backend timeout; slow backends are dropped, not waited on | | DEFAULT_LIMIT | 20 | Default result count | | ELIXA_AFFILIATE_TAG | elixa | Injected as elixa_ref=… query param on outbound URLs |

How ranking works

Every result is scored 0–1 on five axes and a weighted composite decides the final order:

composite =  0.40 · relevance       (keyword match to query)
          +  0.20 · completeness    (0–100, normalised)
          +  0.15 · trust           (Tranco domain rank, normalised)
          +  0.15 · price_fit       (how well it matches price intent)
          +  0.10 · rating          (0–5, normalised)

Weights are defaults. If you're running the server and want different behaviour (e.g. weight trust higher for a luxury use case), fork it — the ranker is deliberately one small file.

Development

npm install
npm run dev       # run from source with tsx
npm run build     # compile to dist/
npm start         # run the compiled build

License

MIT. No warranty. Use at your own discretion.