ERPipe
Enables AI agents to interact with Odoo ERP through natural language, providing tools and prompts for data operations and record management.
README
ERPipe — MCP for Odoo
Open-source TypeScript building blocks for remote MCP access to Odoo.
| This repo (MIT) | Tool logic, transports, single-tenant self-host Worker example |
| Hosted product | Multi-tenant control plane + dashboard — mcp.erpipe.com (separate private repo) |
Try it (ChatGPT-first, real Odoo only): create a free v1 workspace, add one or more HTTPS Odoo instances, then connect ChatGPT Developer Mode once to https://mcp.erpipe.com/mcp. Claude and other MCP clients use the same workspace URL. There is no shared public sandbox.
Why
- Talk to Odoo from ChatGPT (primary), Claude, Cursor, and custom MCP agents over a stable URL shape
- Workers-safe XML-RPC + JSON-2 transports
- Gated writes (preview → approve → execute) with field policy
- Shared 26-tool + 7-prompt core surface in
@erpipe/core(D14 parity set of 23 plusmodel_facts,read_attachment,render_report) — hosted workspace is 37 tools (multi-instance, lifecycle, webhook events,ping) — see docs/tools.md
Status
| Milestone | State |
|---|---|
Hosted workspace /mcp contract |
Implemented in the private cloud control plane |
Self-host /{slug}/mcp example |
Proven — see SPIKE.md |
| Core surface | 26 tools + 7 prompts + 4 resources in @erpipe/core |
| Hosted surface | 37 tools + 7 prompts at mcp.erpipe.com/mcp |
| Pure parity harness | npm run parity — PARITY.md |
| Live smoke (Odoo 16–19) | npm run smoke:live — SMOKE.md |
Pins:
| Package | Version |
|---|---|
agents |
0.17.4 |
@cloudflare/workers-oauth-provider |
0.8.2 |
Packages
| Path | Name | Role |
|---|---|---|
packages/core |
@erpipe/core |
Transports, tools, field policy — no Workers imports |
packages/xmlrpc |
@erpipe/odoo-xmlrpc |
Fetch-based XML-RPC (Workers-safe) |
packages/worker-selfhost |
@erpipe/worker-selfhost |
Single-tenant McpAgent + OAuth example |
Requirements
- Node.js ≥ 20 (CI uses 22 —
.nvmrc) - npm
- Optional: Cloudflare account for
deploy:selfhost - Optional: Odoo 16–19 for live smoke
Install
npm install @erpipe/core @erpipe/odoo-xmlrpc
| Package | npm |
|---|---|
@erpipe/core |
npmjs.com/package/@erpipe/core |
@erpipe/odoo-xmlrpc |
npmjs.com/package/@erpipe/odoo-xmlrpc |
Self-host Worker example stays in this monorepo only (@erpipe/worker-selfhost is private / not published).
Develop from source
git clone https://github.com/tuanle96/erpipe.git
cd erpipe
npm install
npm run build
npm run typecheck
npm test
Quick start (self-host Worker)
npm run dev:selfhost
# Home: http://127.0.0.1:8787/
# MCP: http://127.0.0.1:8787/default/mcp
Deploy:
# set secrets first — see packages/worker-selfhost/README.md
npm run deploy:selfhost
Canonical MCP URLs
| Surface | URL shape | Notes |
|---|---|---|
| Hosted product | https://mcp.erpipe.com/mcp |
One workspace OAuth app; Odoo-bound tools require an explicit instance key (list_instances to discover). |
| Self-host example | https://<host>/{connection_slug}/mcp |
Single-tenant Worker demo in @erpipe/worker-selfhost. |
Hosted setup: mcp.erpipe.com/docs/connector · compatibility: erpipe.com/compatibility
Reserved path segments (never use as a self-host connection id): authorize, token, register, mcp, sse, .well-known, assets, health, app, admin.
Docs
| Doc | Topic |
|---|---|
| docs/tools.md | Full tool & prompt catalog |
| CONTRIBUTING.md | Dev setup & PRs |
| SECURITY.md | Vulnerability reporting |
| CHANGELOG.md | Releases |
| SMOKE.md | Live Odoo smoke |
| PARITY.md | Python vs TS pure harness |
| SPIKE.md | Protocol contract notes |
Architecture (high level)
MCP client ──OAuth + MCP──► Worker (self-host or hosted)
│
▼
@erpipe/core tools
│
┌─────────────┴─────────────┐
▼ ▼
XmlRpcTransport Json2Transport
│ │
└──────────► Odoo ◄─────────┘
Contributing
See CONTRIBUTING.md. Please read the Code of Conduct.
License
MIT — see LICENSE.
"Odoo" is a trademark of Odoo S.A. ERPipe is not affiliated with Odoo S.A.
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。