odoo-mcp
Enables Claude to interact with Odoo through 12 MCP tools covering discovery, CRUD operations, and bulk import/export via XML-RPC.
README
Odoo MCP Multi (own build)
A from-scratch, open MCP server that turns Claude (or any MCP client) into an
operator of the Odoo ORM — exactly the 12 tools described in
Odoo_MCP_Tools_Guide.md.
It is a thin, ergonomic wrapper around Odoo's external API (XML-RPC):
Claude ──(MCP tool)──► server.py ──(XML-RPC /xmlrpc/2/object)──► Odoo (PostgreSQL)
◄──(structured JSON)── ◄──(ORM data)──
Everything ultimately routes through one call:
models.execute_kw(db, uid, key, model, method, args, kwargs).
The 12 tools
| Tool | Group | Risk | Purpose |
|---|---|---|---|
list_available_profiles |
Discover | — | See configured environments (call FIRST) |
get_version |
Discover | — | Odoo server version |
list_models |
Discover | — | Find tables (models) |
list_fields |
Discover | — | See a model's schema |
search_read |
Read | — | Query data (returns pagination envelope) |
get_financial_report |
Read | — | Native reports (Odoo 17+) |
create |
Write | ⚠️ | Create a record |
write |
Write | ⚠️ | Update records |
unlink |
Write | ⚠️⚠️ | Delete records |
execute_kw |
Write | ⚠️⚠️ | Run ANY model method |
export_records |
Bulk | — | Export (backup/migration) |
import_records |
Bulk | ⚠️ | Import/upsert in bulk |
Quick install — one command, no clone (recommended)
Requires uv (pip install uv). This fetches the
code straight from GitHub, installs it in an isolated env, and registers it with
Claude — credentials are passed inline as env vars, so there's no file to edit:
claude mcp add odoo -s user \
-e ODOO_URL=https://your-instance.odoo.com \
-e ODOO_DB=your-db-name \
-e ODOO_USER=you@company.com \
-e ODOO_PASSWORD=your-api-key \
-- uvx --from git+https://github.com/AleemHaider/odoo-mcp odoo-mcp
That's the whole install. Start Claude and ask "List my Odoo profiles".
For multiple Odoo instances, use the profiles.json flow below instead.
Setup (manual / multi-profile)
1. Install
pip install -r requirements.txt # installs fastmcp
2. Configure profiles
Copy the example and fill in your Odoo credentials:
cp profiles.example.json profiles.json
{
"default": "Interhi",
"profiles": {
"Interhi": {
"url": "https://your-instance.odoo.com",
"database": "your-db-name",
"username": "admin",
"password": "your-api-key-or-password"
}
}
}
Tip: In Odoo, generate an API key (Settings → Users → API Keys) and use it as the
password. Mark production profiles with"readonly": trueto block all mutations from that profile.
Alternative (single profile via env vars, no file):
export ODOO_URL="https://your-instance.odoo.com"
export ODOO_DB="your-db-name"
export ODOO_USER="admin"
export ODOO_PASSWORD="your-api-key"
3. Run
python server.py # speaks MCP over stdio
Connect it to Claude
Claude Desktop
Add to claude_desktop_config.json
(macOS: ~/Library/Application Support/Claude/, Windows: %APPDATA%\Claude\):
{
"mcpServers": {
"odoo": {
"command": "python",
"args": ["/absolute/path/to/odoo mcp/server.py"]
}
}
}
Claude Code (CLI)
claude mcp add odoo -- python "/absolute/path/to/odoo mcp/server.py"
Restart the client; the 12 tools appear automatically.
Key concepts (how to drive it)
- Domains use Odoo prefix (Polish) notation — never the words
and/or:["&", ["state","=","posted"], ["move_type","=","out_invoice"]] ["|", ["a","=",1], ["b","=",2]] - Every read returns a pagination envelope:
If{"records": [...], "total": 128, "limit": 100, "offset": 0, "has_more": true, "next_offset": 100, "format": "json"}has_moreis true, call again withoffset = next_offset. - Output
format:json(process),compact(light array-of-arrays),table(Markdown for users),html(paste into Odoo),csv(spreadsheet). - IDs for methods always go in a list:
args=[[490749]]. - Relational command tuples:
[[0,0,{...}]](new o2m line),[[6,0,[id1,id2]]](set m2m). - execute_kw is the swiss-army knife — any Odoo button = a method behind it:
account.move / action_post / [[490749]]validates an invoice.
Safety notes
- Always run
list_available_profilesfirst to confirm the environment. - Read-only introspection via
execute_kw(search/read/fields_get/read_group…) is allowed onreadonlyprofiles; mutating methods are blocked. - Errors come back as
{"success": false, "error": "..."}with the cleaned Odoo message (e.g.Invalid leaf,Expected singleton).
Wraps the Odoo external API. Concept & tool set documented in
Odoo_MCP_Tools_Guide.md.
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。