qsale-mcp
MCP server for the QSale headless commerce platform enabling catalog, content, navigation, segments, and more via a propose→review→apply pattern.
README
qsale-mcp
Model Context Protocol server for the QSale headless commerce platform.
Lets an LLM (Claude Code, Claude Desktop, Cursor, any MCP-compatible client) operate a QSale tenant through the same REST API the admin panel uses: catalog, content pages, navigation, segments, dictionaries, mailings, promotion triggers, redirects, and admin task triggers.
All write operations follow a propose → review → apply pattern: the model first stages the change and returns a before/after diff for you to read; nothing hits the backend until you say so.
Installation
Requires Python 3.11+.
pip install git+https://github.com/qsale-io/qsale-mcp-server.git
Or with uv:
uvx --from git+https://github.com/qsale-io/qsale-mcp-server.git qsale-mcp
Configuration
The server reads its configuration from environment variables — no flags, no config files.
| Variable | Required | Default | Purpose |
|---|---|---|---|
QSALE_API_TOKEN |
yes | — | Employee API token |
QSALE_COMPANY_ID |
yes | — | Company UUID (tenant scope) |
QSALE_API_BASE |
no | https://console.qsale.io |
Base URL of the QSale REST API |
QSALE_CLIENT_TYPE |
no | WEB |
Value of the X-QA-Client-Type header |
Obtaining a token and a company id
- Sign in to your QSale admin panel.
- Open Settings → API tokens and create a new employee token. Save it somewhere safe — it is shown only once.
- The company UUID is visible in the admin URL or under Settings → Company.
Self-hosted installations override QSALE_API_BASE to point at their own
console host (e.g. https://console.example.com).
Wiring into Claude Code
Add to your .mcp.json (or the global Claude Code MCP config):
{
"mcpServers": {
"qsale": {
"command": "qsale-mcp",
"env": {
"QSALE_API_TOKEN": "your-token-here",
"QSALE_COMPANY_ID": "00000000-0000-0000-0000-000000000000"
}
}
}
}
For Claude Desktop, the configuration file lives at
~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
or %APPDATA%\Claude\claude_desktop_config.json (Windows) with the same
mcpServers shape.
Available tools
The server registers ~100 tools, grouped by domain. See the source files
or the model's tools/list output for the full list; below is the high
level shape.
Read
Listing and detail tools for:
- Product categories, products
- Content pages
- Navigation groups and items
- URL redirects and redirect sites
- Mail templates and template images
- Promotion triggers and trigger categories
- Dictionaries and dictionary items
- Segments, segment properties, segment filters, segment property choices
- Frontend settings
Write — propose / apply
Two-phase writes that stage the change in memory, return a diff, and only
hit the backend when apply_* is called. Cover:
propose/apply_page_updatepropose/apply_navigation_item_create + updatepropose/apply_category_create + deletepropose/apply_mail_template_create + updatepropose/apply_promotion_trigger_createpropose/apply_dictionary_create + deletepropose/apply_dictionary_item_create + deletepropose/apply_segment_create + deletepropose/apply_segment_property_create + update + deletepropose/apply_segment_property_choice_create + deletepropose/apply_segment_filter_create + update + deletepropose/apply_link_di_segment + unlink_di_segmentpropose/apply_link_pc_segment + unlink_pc_segment
Admin task triggers
Schedule asynchronous backend tasks via the admin task endpoints:
propose/apply_run_update_all_dictspropose/apply_run_update_dictpropose/apply_run_set_category_for_products
Direct writes (no two-phase)
Idempotent or low-impact operations are exposed as single calls:
create_redirect,update_redirect,create_redirect_site,update_redirect_siteupdate_categoryupdate_frontend_setting_json,set_frontend_setting_filecreate_mail_template_image
Batch
bulk_apply([proposal_ids])— one approval covers N already-staged proposals of any kind. Stops on the first failure and returns per-proposal results.
How writes work
propose_* ──► in-memory Proposal{ id, kind, fields, before } (no HTTP write)
│
▼
you read the diff in chat
│
▼
apply_* ──► REST POST/PATCH/DELETE on console.qsale.io
Proposals live only in process memory — restarting the MCP server discards them. This is intentional: a stale proposal should never be reused after a code reload.
Development
git clone https://github.com/qsale-io/qsale-mcp-server.git
cd qsale-mcp-server
python -m venv .venv && source .venv/bin/activate
pip install -e '.[dev]'
ruff check .
pytest
Versioning
This project follows Semantic Versioning. See CHANGELOG.md for release notes.
Security
Found a vulnerability? See SECURITY.md. Do not open a public issue for security reports.
License
Apache License 2.0 — see 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。