Final POS MCP
Enables AI agents to build, preview, validate, and deploy custom point-of-sale, kiosk, self-checkout, KDS, and in-person payment applications using Final's platform.
README
<p align="center"> <img src="assets/final-icon.png" alt="Final" width="128" /> </p>
<h1 align="center">Final POS MCP</h1>
<p align="center"> Build custom <strong>point of sale, kiosk, self-checkout, KDS & in-person payment apps</strong> with AI agents —<br /> from Gemini CLI, Claude Code, Cursor, ChatGPT, or any MCP client. </p>
<p align="center"> <a href="https://finalpos.com">finalpos.com</a> · <a href="https://build.finalpos.com">Final Builder</a> · <a href="https://docs.finalpos.com">API Docs</a> </p>
Final lets anyone create their own POS — every pixel, every permission, every interaction — and run it on any hardware: desktop, tablet, mobile, kiosk, payment terminal, line buster, or menu board, with integrated card payments in 24+ countries.
This repo is the official home for connecting your own AI agent to Final. It ships:
- A Gemini CLI extension (this repo installs directly).
- The hosted Final Flows MCP server — 20 tools for grounded API lookups, vetted screen blueprints, flow file writes, live sandbox previews, validation gates, and deploys.
- The Final docs MCP server — search the Final POS REST API.
- Setup snippets for Claude Code, Cursor, ChatGPT, and any MCP client.
Everything your agent builds appears live in the Final builder: the app shows up in your home grid, the agent's session appears as a chat, and every file write is a restorable checkpoint.
Get an API key
All clients authenticate with a Final organization API key (sk_...):
- Sign in at build.finalpos.com.
- Open a flow, then in the chat rail click the cable icon → "Connect your AI tool".
- Mint a key — it is shown once, so copy it right away.
Install
Gemini CLI
gemini extensions install https://github.com/Final-Commerce/final-pos-mcp
You'll be prompted for your FINAL_API_KEY during install (stored in your
system keychain). Change it later with:
gemini extensions config final-pos
Then start building:
/final-pos:new a coffee-shop kiosk with a product grid, cart, and card payments
/final-pos:preview # live sandbox preview URL
/final-pos:validate # run the full gate suite
/final-pos:deploy # ship it
Claude Code
claude mcp add final-flows --transport http https://build.finalpos.com/api/mcp \
--header "Authorization: Bearer sk_YOUR_KEY"
Cursor
Or add to ~/.cursor/mcp.json manually:
{
"mcpServers": {
"final-flows": {
"url": "https://build.finalpos.com/api/mcp",
"headers": { "Authorization": "Bearer sk_YOUR_KEY" }
}
}
}
ChatGPT and any other MCP client
The server speaks standard Streamable HTTP:
- Endpoint:
https://build.finalpos.com/api/mcp - Auth header:
Authorization: Bearer sk_YOUR_KEY(orx-api-key: sk_YOUR_KEY)
In ChatGPT, add it as a connector under Settings → Connectors (developer mode). Works the same in VS Code, Windsurf, Zed, and anything else that speaks MCP.
What your agent can do
| Category | Tools |
|---|---|
| Grounding | lookup_command_frame, lookup_pos_api, lookup_engine_component, list_blueprints, get_blueprint, get_scaffold |
| Flow lifecycle | list_flows, create_flow, connect_chat, log_message, read_flow_files, write_flow_files, delete_flow_files |
| Preview & ship | start_preview, sync_preview, get_preview_status, stop_preview, validate_flow, open_in_builder, deploy_flow |
A typical build: get_scaffold → create_flow → ground every UI call in
lookup_command_frame → adapt a vetted blueprint → write_flow_files →
validate_flow → start_preview → hand you a live URL.
All tool calls are scoped to your own organization's flows. Rate limits per key: 30 preview boots/hour, 12 builds+deploys/hour, 300 file writes/hour.
Security
- Your API key is org-scoped and revocable; the secret is never stored by this extension repo — Gemini CLI keeps it in your system keychain.
- The MCP servers are hosted by Final; no code from this repo runs locally except prompt/context files.
- Agents cannot touch engine-owned files, and every deploy passes guardrail, typecheck, and secret scans.
License
Apache-2.0 © Final Commerce
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。