Let your agent shop and compare: Shopify catalog search, offer ranking, shipping plans. $0.01/call.
Runs on your machine from its package; the steps are below.
Not assessed yet.
Odel charges nothing per call. The publisher may charge for its own service.
Runs locally over stdio
This server isn't hosted — your MCP client launches it from a package registry. Use one of the commands below, or drop the config into your client (e.g. Claude Desktop).
Run
npx -y @forgemeshlabs/shopscout-mcp
MCP client config
{
"mcpServers": {
"shopscout": {
"command": "npx",
"args": [
"-y",
"@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.
Requires Node.js 20 or later.
{
"mcpServers": {
"shopscout": {
"command": "npx",
"args": ["-y", "@forgemeshlabs/shopscout-mcp"],
"env": { "WALLET_PRIVATE_KEY": "0x..." }
}
}
}
WALLET_PRIVATE_KEY 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, get_capabilities still works and every paid tool returns the x402 challenge as data (payment_required) 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 X402_MAX_PRICE_USD and X402_SESSION_BUDGET_USD can only lower those caps, never raise them.
| Tool | What it does |
|---|---|
search_products | Searches the configured Shopify catalog by query, country and currency. |
get_product | Refreshes a product/variant and selected options. |
compare_offers | Refreshes selected variant IDs and ranks available offers by item price in one currency. |
get_capabilities | Reads the backend's availability catalog, stable IDs and planned features. |
compare_shipping_plans | 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 SHOPSCOUT_CATALOG_ENABLED=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. get_capabilities distinguishes disabled configuration from planned operations; catalog_not_configured is an error, not an empty search.
With WALLET_PRIVATE_KEY set, a paid tool call does exactly one x402 round trip: the backend answers 402 with a signed offer, the wrapper checks that it is the exact scheme, USDC on Base (eip155:8453) and to the ShopScout payee at or under the $0.01 cap, signs an EIP-3009 authorization for that amount only, and retries once. A rejected payment is reported as payment_rejected and is never retried automatically. Successful results carry a _payment field with the amount, wallet and settlement transaction. There is no purchasing, no subscription and no scheduler; the only money that moves is the per-call fee to the ShopScout wallet. The signer reads no .env file; pass the key through your MCP client's env block.
Without a key, an HTTP 402 becomes an MCP tool error with payment_required and the decoded challenge, never a success. Catalog data is returned as data, never as instructions.
npm run build
SHOPSCOUT_SERVER_PATH=/absolute/path/to/x402-shopscout-server npm test
The integration test starts an ephemeral backend, connects through a real MCP SDK stdio client, lists all five tools and exercises search, detail lookup, offer comparison and shipping with synthetic catalog data. No production API, merchant catalog or payment is called. Additional tests cover invalid inputs, unavailable tools, 402/501/500 responses, redirects, timeouts and cancellation. Tests require permission to bind localhost.
contracts/openapi.json is a pinned backend contract, not a second implementation. To update after a backend contract change:
node /absolute/path/to/x402-shopscout-server/scripts/export-contracts.js
cp /absolute/path/to/x402-shopscout-server/openapi.json contracts/openapi.json
npm run build
Review new operations deliberately; do not automatically expose planned endpoints. npm run build compiles JSON schemas and generates tool-manifest.json; source JavaScript runs directly.
Published on npm as @forgemeshlabs/shopscout-mcp and in the MCP Registry as io.github.forgemeshlabs/shopscout-mcp. Source: https://github.com/forgemeshlabs/shopscout-mcp. Product page and pricing: https://forgemesh.io/shopscout. The Dockerfile builds the same stdio server for container use.