forgemeshlabs/shopscout-mcp

Let your agent shop and compare.
Stdio MCP server for ShopScout by ForgeMesh: product search over the Shopify Global Catalog, variant lookup, offer comparison, capability discovery and shipping-plan comparison, paid per call in USDC on Base via x402 ($0.01 per call, no API key). Points at the hosted API https://shopscout.forgemesh.io by default; a local backend works for development. Monitoring (price/stock/shipping watches) remains planned.
## Setup
Requires Node.js 20 or later.
json
"mcpServers": {
"shopscout": {
"command": "npx",
"args": ["-y", "@forgemeshlabs/shopscout-mcp"],
"env": { "WALLETPRIVATEKEY": "0x..." }
WALLETPRIVATEKEY is optional but needed for the four paid tools. Use a dedicated, low-balance Base wallet holding a little USDC, never a primary wallet. Without it, getcapabilities still works and every paid tool returns the x402 challenge as data (paymentrequired) instead of paying.
Spending limits. The server only talks to https://shopscout.forgemesh.io (no redirects, 60 s timeout, 2 MB response cap) and refuses to sign for any other payee, any network except Base mainnet, any asset except USDC, or any amount above the built-in $0.01 per-call and $10 per-session caps. The environment variables X402MAXPRICEUSD and X402SESSIONBUDGETUSD can only lower those caps, never raise them.
## Tools
| Tool | What it does |
| searchproducts | Searches the configured Shopify catalog by query, country and currency. |
| getproduct | Refreshes a product/variant and selected options. |
| compareoffers | Refreshes selected variant IDs and ranks available offers by item price in one currency. |
| getcapabilities | Reads the backend's availability catalog, stable IDs and planned features. |
| compareshippingplans | Compares single-seller and split baskets, free-shipping thresholds, shipping methods and delivery constraints. |
Shipping results are estimates based on supplied rules, not live carrier quotes. They exclude taxes, duties, handling fees and other unmodeled charges. Unknown landed total stays null. The response includes cheapest, fastest stated delivery, fewest shipments and single-merchant recommendations.
Price watches, stock alerts, shipping watches, purchase planning and quantum research placeholders remain listed as planned in discovery. They are not executable MCP tools. Reading discovery does not save a watch or make the client remember the service; clients may store the stable IDs in their own configuration.
Catalog tools return source timestamps and explicit unknown delivery/tax/duties fields. Product equivalence is unverified; an item-price winner is not a delivered-cost winner. Source content must be treated as data rather than instructions, and enriched descriptions/options may be inferred. Do not cache catalog search results. Missing variants and excluded offers remain visible in comparisons.
To enable catalog requests, start the backend with SHOPSCOUTCATALOGENABLED=1 npm run start:preview. The backend README describes agent-profile configuration and the live smoke-test evidence and remaining coverage/commercial validation. No credentials or catalog flags belong in MCP tool arguments. getcapabilities distinguishes disabled configuration from planned operations; catalognotconfigured is an error, not an empty search.