NorthCinder
NorthCinder is an MCP server that enables AI agents to compare products against a brief, rank them with explanations, and support approved checkout, all while running locally with privacy controls.
README
NorthCinder
Make your shopping agent compare, explain, and ask before buying.
NorthCinder is an open-source MCP server that helps an AI agent compare products against your brief. It returns a ranked shortlist with reasons, reports which stores it could and could not search, and requires a separate approval before checkout. Seller payment and affiliate data never improve a result's position.
NorthCinder is software you run. It works on your computer alongside the AI app you already use. The repository owner does not operate a NorthCinder service. There is no NorthCinder account, hosted control plane, or telemetry service.
Get started
You need Node.js 20 or later and an AI app that supports MCP.
npx northcinder init
The initializer walks you through local setup and prints the MCP configuration for your AI app. After you connect it, try a specific brief:
Find black wool running shoes under $130. Compare price, delivery, fit, and merchant trust. Tell me why the winner ranked first and which options were ruled out.
What you get
NorthCinder's response includes:
- the offers that matched the required criteria;
- the score and machine-readable reasons for each recommendation;
- rejected offers and the requirement each one missed;
- merchant-trust evidence and missing evidence;
- a store-by-store coverage report; and
- sponsorship and source labels that remain attached to every offer.
The contract you can inspect
| Concern | NorthCinder's rule | Evidence |
|---|---|---|
| Ranking | Buyer criteria determine the order. Seller payment is not an input. | Ranking specification and ranking source |
| Sponsored offers | Labeled sponsored offers remain below every organic result. | Neutrality audit |
| Coverage | Unavailable and unconfigured stores stay visible in the response. | Adapter contract |
| Merchant trust | Every merchant carries explicit evidence or an honest unknown state. | Trust specification |
| Checkout | A signed, single-use mandate binds the exact offer, quantity, and spending cap. | Checkout package |
| Audit | Recommendations, approvals, and checkout attempts are written to a local audit trail. | Client source |
The client reruns the deterministic ranking over the service-disclosed inputs. This verifies the order it received; it does not prove that an upstream catalog was complete or that every store-supplied fact was true.
How NorthCinder works
flowchart LR
A["Your AI app"] -->|MCP| C["NorthCinder client"]
C --> S["Buyer-run search service"]
S --> D["Configured store adapters"]
S --> C
C --> R["Local reranking and reasons"]
C --> L["Local audit trail"]
C --> P["Buyer approval"]
P -->|signed single-use mandate| X["Checkout rail or cart handoff"]
The repository owner is not in this runtime path. You run the client and aggregation engine, choose the store connections, and keep the local configuration and audit data.
Store coverage
Store access varies because each platform has different rules. NorthCinder reports a store as unavailable or blocked when it cannot search it instead of presenting partial coverage as a complete market search.
| Store | What works today |
|---|---|
| Shopify | Catalog search is available after additional Shopify setup. The older per-store connection is no longer current. |
| WooCommerce | Works with stores that expose WooCommerce's public Store API. |
| eBay | Native search requires approved eBay Buy API access. |
| Etsy | Native search requires approved Etsy app access. |
| Amazon | Read-only comparison can use a browser profile you control. It stops at challenges and does not check out. |
When a native store connection is unavailable, your agent may still compare permitted product pages with browser tools it already controls. NorthCinder accepts normalized product facts, not cookies, raw HTML, screenshots, page instructions, passwords, one-time codes, or your AI-provider key. Agent-observed offers must be confirmed by a native adapter or merchant protocol before automated checkout or an unattended watch.
Privacy and purchase control
- Your AI-provider key stays in your AI app. Store credentials stay with you and the store.
- NorthCinder does not send your searches, settings, or local history to the repository owner.
- Raw card details are rejected. Automated rails use opaque delegated payment tokens when available.
- Search and price watches are not permission to buy. Every automated checkout needs approval for that specific purchase.
- The approval mandate is single use and protected by a cross-process nonce ledger.
Read privacy and software ownership and the security policy for the complete boundary.
Project status
northcinder 0.1.2 is the initial public release. The repository includes offline build, typecheck, test,
dependency, secret, packaging, and public-surface checks. Fixture and harness coverage does not prove current
third-party credentials, production access to every store, or a completed real purchase. Read the relevant
adapter documentation before relying on a specific integration.
The package is not yet listed in the official MCP Registry, and the marketing site in this repository has not been published at a canonical production origin.
Build from source
This is a pnpm workspace. Product packages require Node.js 20 or later; the private site workspace requires Node.js 22.12 or later.
corepack pnpm install --frozen-lockfile
corepack pnpm build
Then run the local initializer from the checkout:
node northcinder/bin/northcinder.js init
Contributors can find the full release-verification commands in CONTRIBUTING.md.
Contributing and support
- Read CONTRIBUTING.md before proposing a change.
- Use GitHub Issues for reproducible bugs and focused work.
- Use GitHub Discussions for questions and open-ended ideas.
- Report vulnerabilities through GitHub private vulnerability reporting, not a public issue.
NorthCinder is open source under the MIT License.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。