MCP Server: Connect Claude, ChatGPT and AI Agents to Odoo

Instead of exporting to Excel and pasting into a chat, connect your AI assistant to Odoo and ask in plain words:

• "How did we sell last week, compared with the week before?"
• "Which customers are more than 30 days late? Draft the reminders."
• "Make a quotation from this email."
• "What should we reorder, and from which vendor?"
• "Where is the pipeline stuck this month?"

The assistant reads live data with links back to each record, and prepares quotations, invoices, requests for quotation, leads and tasks as drafts.

The app adds a Model Context Protocol endpoint to your database, at an address such as https://your-odoo/sqx_mcp. There is no separate server or Docker container to host, and no Python library to install. It follows the current MCP specification with the Streamable HTTP transport, and uses the Odoo ORM directly rather than the XML-RPC and JSON-RPC APIs that Odoo has deprecated. It works on Community, Enterprise and Odoo.sh.

• Claude and ChatGPT: add a custom connector with the server address. A window opens on your Odoo: the user signs in with OAuth 2.1, chooses read only or read and change, and approves. Connectors added on the web also work in the desktop and mobile apps.
• Claude Code: create an API key in your Odoo preferences, then add the server with claude mcp add --transport http and an Authorization header.
• Cursor and other editors: add the address and the key to mcp.json.
• n8n and AI agents: use the address as a Streamable HTTP endpoint with the API key as a Bearer header.

An API key can be restricted to this server only, so it never opens XML-RPC or JSON-RPC.

Generic tools search, count, read and group any model. Business tools answer the usual questions directly: a sales or purchase summary, overdue invoices and the aged balance, low stock and availability per warehouse, the pipeline by stage, open tasks by deadline. Writing tools create draft quotations priced by Odoo, draft invoices and vendor bills that are never posted, requests for quotation, leads and tasks. A tool only shows up when its app is installed, and each one can be switched off or kept for a group.

• Every call runs as the user who approved the connection, with their access rights, record rules and companies. There is no shared technical account.
• Read only by default; creating or changing records is a separate permission, and each change is previewed before it is saved.
• Passwords, tokens, settings and technical models are never exposed.
• Every call is logged with the user, the assistant, the tool and its duration, each changed record gets a note in its chatter, and each connection can be revoked in one click.