odoo-mcp
Enables AI agents to interact with Odoo 19 via JSON-2 API, supporting CRUD operations on Odoo models through MCP tools.
README
odoo-mcp
Docker-first FastMCP server for Odoo 19 JSON-2 integrations.
odoo-mcp exposes Odoo as an MCP server over both HTTP and stdio. It uses the
Odoo 19 external JSON-2 API only:
POST /json/2/<model>/<method>
Authorization: bearer <ODOO_API_KEY>
X-Odoo-Database: <optional database>
Legacy XML-RPC and JSON-RPC are intentionally out of scope.
Status
This is an early implementation with a strong v1 foundation:
- FastMCP
3.2.4 - Docker Compose first
- HTTP and stdio transports
- lazy Odoo credential loading so MCP discovery works before secrets are present
- guarded write operations with
confirm=True - generic Odoo tools
- comprehensive CRM tool pack
- resources and prompts for safe Odoo work
- pytest, ruff, mypy, and GitHub Actions CI
Quick Start With Docker
Copy the environment template and edit the Odoo values:
cp .env.example .env
Required for real Odoo calls:
ODOO_URL=https://your-odoo-host.example.com
ODOO_API_KEY=your-odoo-api-key
ODOO_DATABASE=your-database-if-needed
Start the HTTP server:
docker compose up --build
The default MCP endpoint is:
http://localhost:8000/mcp
Check local process readiness:
docker compose exec odoo-mcp uv run --no-sync odoo-mcp healthcheck
Local Development
Install dependencies:
uv sync
Run the full gate:
uv run pytest
uv run ruff check .
uv run ruff format --check .
uv run mypy src
Run HTTP locally:
uv run odoo-mcp http --host 0.0.0.0 --port 8000 --path /mcp
Run stdio locally:
uv run odoo-mcp stdio
Inspect sanitized config:
uv run odoo-mcp inspect-config
MCP Client Config
Stdio example:
{
"mcpServers": {
"odoo": {
"command": "uv",
"args": ["run", "odoo-mcp", "stdio"],
"env": {
"ODOO_URL": "https://your-odoo-host.example.com",
"ODOO_API_KEY": "your-odoo-api-key",
"ODOO_DATABASE": "your-database-if-needed"
}
}
}
}
HTTP clients should connect to /mcp. If the server is exposed beyond a trusted
network, set MCP_AUTH_MODE=static or MCP_AUTH_MODE=jwt.
Configuration
| Variable | Purpose |
|---|---|
ODOO_URL |
Base URL for Odoo 19 JSON-2 calls. |
ODOO_API_KEY |
Odoo API key sent as Authorization: bearer .... |
ODOO_DATABASE |
Optional Odoo database header. |
ODOO_TIMEOUT_SECONDS |
HTTP timeout for Odoo calls. Defaults to 30. |
ODOO_ALLOWED_GENERIC_METHODS |
Comma-separated allowlist for otherwise blocked generic methods. |
ODOO_CRM_OPTIONAL_FEATURES |
Comma-separated optional CRM features to expose. |
MCP_AUTH_MODE |
none, static, or jwt. Defaults to none. |
MCP_STATIC_TOKEN |
Bearer token for MCP_AUTH_MODE=static. |
JWT_JWKS_URI |
JWKS URL for MCP_AUTH_MODE=jwt. |
JWT_PUBLIC_KEY |
Public key alternative for JWT verification. |
JWT_ISSUER |
Optional expected JWT issuer. |
JWT_AUDIENCE |
Optional expected JWT audience. |
JWT_REQUIRED_SCOPES |
Optional comma-separated JWT scopes. |
MCP_HTTP_HOST |
HTTP bind host. Defaults to 0.0.0.0. |
MCP_HTTP_PORT |
HTTP bind port. Defaults to 8000. |
MCP_HTTP_PATH |
MCP endpoint path. Defaults to /mcp. |
MCP_LOG_LEVEL |
HTTP server log level. Defaults to info. |
Tool Catalog
Generic Odoo tools:
odoo_search_readodoo_readodoo_createodoo_writeodoo_unlinkodoo_actionodoo_call_methododoo_current_userodoo_model_fieldsodoo_list_models
CRM tools:
- leads and opportunities:
crm_list_leads,crm_get_lead,crm_create_lead,crm_update_lead,crm_assign_lead - pipeline:
crm_list_pipeline_stages,crm_move_lead_to_stage,crm_pipeline_report - won/lost lifecycle:
crm_mark_won,crm_mark_lost,crm_restore_lead,crm_list_lost_reasons - conversion and merge:
crm_convert_lead_to_opportunity,crm_merge_opportunities - activities:
crm_schedule_activity,crm_mark_activity_done,crm_list_activities,crm_list_activity_types,crm_activity_report - teams and scoring:
crm_list_teams,crm_update_lead_score - optional features:
crm_enrich_lead,crm_list_scoring_rules,crm_recurring_revenue_report
Optional CRM tools are registered only when their feature key is listed in
ODOO_CRM_OPTIONAL_FEATURES.
Resources And Prompts
Resources:
odoo://server/infoodoo://user/contextodoo://model/{model}/fieldsodoo://crm/pipeline/summary
Prompts:
odoo_safe_operationodoo_crm_pipeline_reviewodoo_record_change_plan
Safety Model
Odoo remains the final authorization layer through its access rights and record rules. The MCP server adds local guardrails:
- mutating Odoo client calls require
confirm=True - generic dangerous methods such as
call_kwandexecute_kware blocked unless explicitly allowlisted - HTTP MCP auth is separate from the server-side Odoo API key
inspect-configredacts secrets- Odoo API errors redact configured secret values
Use MCP_AUTH_MODE=static for simple private HTTP deployments:
MCP_AUTH_MODE=static
MCP_STATIC_TOKEN=change-me
Use MCP_AUTH_MODE=jwt with JWT_JWKS_URI or JWT_PUBLIC_KEY for production
identity-provider-backed deployments.
Extending
Add module packs under src/odoo_mcp/tools/ and register them from
src/odoo_mcp/server.py. Prefer typed, curated tools for business workflows and
leave odoo_call_method as the explicit escape hatch.
When adding new Odoo workflows:
- Write tests first with mocked Odoo JSON-2 behavior.
- Prefer single-call Odoo methods like
search_readbecause each JSON-2 call is its own transaction. - Require
confirm=Truefor writes, actions, posting, validation, archive, and delete operations. - Add focused README entries for new tools and optional feature flags.
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。