northcinder

Why we included this project

Teams building shopping or e-commerce agents often hit the same wall: a model will happily recommend whatever a search returns, and buyers have no way to verify the order. NorthCinder addresses that by running as an MCP server you install locally, giving your agent a deterministic, buyer-criteria-driven comparison workflow across stores like Shopify, WooCommerce, eBay, Etsy, and Amazon. Every ranked result carries machine-readable reasons, rejected offers show which requirement they missed, and merchant-trust evidence travels with each merchant. The design is deliberately adversarial to ad influence — seller payments and affiliate data never feed into ranking, sponsored offers stay below organic results, and the client independently reruns the ranking to verify the order it received. It also adds a hard human-in-the-loop guardrail: checkout requires a signed, single-use mandate tied to a specific offer, quantity, and spending cap, with everything written to a local audit trail. That makes it a practical choice for anyone who wants an auditable, privacy-respecting shopping agent instead of a black-box recommender.