RAG Odoo MCP Server - Claude & AI Connector for Odoo

Connect Claude & Compatible AI Clients With Odoo Through MCP

RAG Odoo MCP Server runs inside Odoo and exposes selected business capabilities through the Model Context Protocol. AI clients can discover accessible Odoo models, search approved records, retrieve individual records, create authorized records and inspect server/connection information.

The product also includes configurable Odoo knowledge records that teams can use to store guidance and context for AI-assisted workflows.

Important: What “RAG” Means in This Product

This product does not use a vector database, embeddings or automatic semantic indexing. Its knowledge layer is based on manually created Odoo knowledge snippets that can be organized by category, model, company, tags, priority and an always-include setting. Live Odoo business data is queried using MCP tools at request time.

Architecture fit

If your project specifically requires vector embeddings, semantic retrieval, document chunking or automatic indexing, those capabilities are not included in this module.

Key Features

• Supports Odoo 10-19 with separate compatible packages per series
• Supports Odoo Community and Enterprise
• Designed primarily for Claude Desktop and Claude Mobile
• Cursor and ChatGPT agents are named as compatible MCP clients
• Streamable HTTP JSON-RPC 2.0 on POST /mcp
• Bearer token, X-MCP-API-Key and OAuth 2.0 with PKCE
• Five MCP tools for models, record search, record retrieval, record creation and server information
• MCP calls run as the linked Odoo user, so Odoo ACLs and record rules apply
• Additional model, field and domain restrictions
• Manual knowledge snippets for guidance/context
• Audit logging
• Direct AI-client-to-Odoo communication with no vendor relay server
• No external telemetry and no AI-provider key required inside Odoo
• No Redis, vector database or separate MCP daemon required
• Full source code included

Supported Odoo Versions & Editions

Separate addon builds are available for Odoo 10, 11, 12, 13, 14, 15, 16, 17, 18 and 19. Both Community and Enterprise editions are supported.

Version delivery

Use the package that matches the buyer’s Odoo series. If your WebbyTemplate checkout process uses version confirmation after purchase, ask the buyer to email the required Odoo version before delivery.

Supported AI Clients

Primary Documented Clients

• Claude Desktop
• Claude Mobile

Other Named MCP Clients

• Cursor
• ChatGPT agents
• Other compatible MCP clients

Claude is the primary documented target. Compatibility with other clients depends on their MCP and authentication support; the addon source does not include an automated cross-client test suite.

MCP Protocol & Transport

The module uses Streamable HTTP with JSON-RPC 2.0. The primary endpoint is POST /mcp and the implementation identifies MCP protocol version 2024-11-05. It also provides a legacy SSE handshake endpoint, but long-lived SSE transport is not implemented.

Primary endpoint: POST /mcp

For multi-database Odoo environments, include the database identifier in the connection URL where required.

Available MCP Tools

list_models: Lists Odoo models available to the connected token.

search_records: Searches accessible Odoo records using Odoo domains and permitted fields.

get_record: Retrieves a specific accessible record.

create_record: Creates a new Odoo record when the connected token is permitted to do so.

server_info: Returns connection information such as database, user, company, token and access mode.

Read & Write Capabilities

The current MCP tool layer supports read operations and record creation. It does not expose dedicated update_record or delete_record tools. Even when a token configuration contains broader CRUD flags, the MCP tool set itself does not expose update or delete operations.

How Odoo Permissions Are Applied

MCP calls run as the Odoo user linked to the token, so Odoo ACLs, record rules and user permissions remain part of the access model. Additional token controls can restrict allowed models, read/create/write/unlink flags, fields and domains. In custom access mode, a model without an explicit permission line is denied.

Knowledge Snippets

The module includes a rag.mcp.knowledge model for storing human-created guidance snippets. Knowledge can be organized by category, model, company, tags, priority and an always-include setting. These records are not embedding vectors.

Current source behavior

The addon contains knowledge-retrieval logic, but source verification found that these knowledge snippets are not automatically injected into the MCP initialize/tool flow. Treat them as configurable guidance stored in Odoo, not as an automatic vector-RAG pipeline.

Search Behaviour

Record search uses standard Odoo domain search rather than semantic or hybrid retrieval. The default record-search limit is 20 with a configured maximum of 200. The module does not provide vector similarity scores or a similarity threshold.

Document Jobs

The module can associate Odoo attachments with document-job records. It does not include its own PDF, DOCX, spreadsheet or image parser. Content extraction is expected to be performed by the connected AI client. This is a workflow helper rather than an automatic document-indexing pipeline.

Authentication

Bearer Token

Standard bearer authentication is implemented.

X-MCP-API-Key

API-key-style header authentication is implemented.

OAuth 2.0 + PKCE

OAuth authorization-code flow with PKCE is implemented.

Client Credentials

OAuth client-credentials flow is also implemented.

For Claude connectors, the Odoo MCP endpoint should be available over public HTTPS.

Data Privacy

Data path: AI Client ↔ Your Odoo Instance

There is no WebbyCrown relay server, no required vendor middleware and no external telemetry. AI-provider API keys are not required inside Odoo; the connected client is responsible for its own AI subscription, credentials and model usage.

Audit Logs

The addon records MCP activity including method, tool, payload, response summary, duration, status, IP and request ID. Default audit retention is 30 days. A purge method exists in source, but the current addon does not install an automatic scheduled cron for that purge; teams requiring automatic retention enforcement should configure or review this before production use.

Multi-Company Notes

The product contains company-related fields and token company lists, but current source verification found that token company_ids are not explicitly applied as an additional domain inside search_records. Odoo user ACLs and record rules therefore remain especially important in multi-company deployments.

Deployment

• Self-hosted Odoo
• Odoo.sh
• Docker-based deployments

A publicly reachable HTTPS endpoint is required for Claude connector use. No Redis service or separate MCP daemon is required.

No Vector Database or Embedding Service Required

The module does not depend on Qdrant, pgvector, Chroma, FAISS, OpenAI embeddings, Ollama embeddings or another vector database. This keeps infrastructure requirements lighter, but it also means this product is not a fit when embedding-based semantic RAG is a requirement.

Security Implementation Notes

Bearer secrets are hashed using SHA-256 for lookup. The current implementation also retains a plain-token value so OAuth can re-issue the associated MCP bearer token. For production environments, review Odoo access to token records, HTTPS configuration, linked-user privileges, model rules, field restrictions, domain rules and OAuth clients. The current addon does not include a built-in IP allowlist or rate limiter.

Common Use Cases

Claude + Odoo Record Access: Allow Claude to search and retrieve approved Odoo business records.

AI-Assisted CRM: Let an authorized AI client search CRM data or create approved records.

Internal Odoo Assistant: Expose selected Odoo models to an MCP-compatible company assistant.

Controlled Knowledge Guidance: Store business or workflow instructions as Odoo knowledge snippets.

Document Intake Workflows: Associate documents with Odoo jobs and allow the AI client to extract and process content.

Custom MCP Development: Use the full source code as a starting point for extending Odoo-specific MCP workflows.

What Is Included?

• Full Python/XML source code
• Security groups and Odoo access configuration
• Settings and demo data
• Sample tokens and knowledge records
• Dashboard data and document-job examples
• Sample audit data
• Product/store assets

What Is Not Included?

• Claude, ChatGPT or Cursor subscription
• AI-provider usage fees
• Hosting and SSL/tunnel services
• Odoo license
• Vector database
• Embedding API
• Ollama
• Separate MCP daemon

Known Limitations

• No vector embeddings or vector database.
• No semantic/hybrid retrieval or automatic document indexing.
• No document chunking or automatic re-indexing.
• No dedicated update_record or delete_record MCP tools.
• No MCP Resources, Prompts or Sampling.
• No durable long-lived SSE transport.
• No built-in rate limiting or IP allowlisting.
• Audit-purge method exists but automatic purge cron is not installed.
• No language-aware retrieval layer.