1. Prerequisites
- Odoo 17.0 Community or Enterprise.
- PostgreSQL as the Odoo database.
- Administrator access to Odoo Settings.
- Python package
asyncpg installed in the Odoo Python environment for embedded mode.
- Python package
openpyxl installed for Excel export.
- HTTPS for production deployments.
Optional AI Chat dependencies are openai for OpenAI-compatible endpoints and
anthropic for Anthropic-compatible endpoints. MCP access does not require OpenAI,
Anthropic, or a cloud LLM provider.
2. Install the Addon
- Install the required Python packages in the same environment that runs Odoo.
- Copy or deploy the addon into the Odoo addons path.
- Update the Odoo app list.
- Install Foggy Odoo Bridge Pro.
- Restart Odoo after installation if your deployment process does not restart it automatically.
3. Configure the Engine
- Open Settings -> Foggy MCP.
- Keep Engine Mode set to Embedded for the recommended self-contained setup.
- Run the health check.
- Confirm that the embedded engine, required model files, and required Python packages are available.
Gateway mode remains available for advanced deployments that already run a separate Foggy service.
4. Create API Keys for MCP Clients
- Open Foggy MCP -> API Keys.
- Create an API key for each user or integration that needs MCP access.
- Store the generated key securely. It is shown only once.
- Connect your MCP client to
https://your-odoo-domain.example/foggy-mcp/rpc.
Use the key as a bearer token:
Authorization: Bearer fmcp_xxx
Access is evaluated as the Odoo user that owns the API key. Odoo record rules, company access,
and Foggy column governance are enforced server-side.
5. Optional Built-In AI Chat
Built-in AI Chat is optional. MCP access works without it.
- Open Settings -> Foggy MCP -> AI Chat Configuration.
- Select a provider.
- Configure the model, API key, and optional base URL.
- Review your provider and data boundary.
- Enable AI Chat LLM Data Transfer Consent.
Before this consent is enabled, built-in AI Chat will not send prompts or selected query context
to the configured LLM endpoint.
6. AI Data Boundary
AI Chat may send the user's prompt, available model metadata, selected query context, and query results needed for the answer.
If you use a self-hosted OpenAI-compatible endpoint, Ollama, or a private gateway, LLM traffic can
remain inside your controlled boundary. If you use a cloud provider such as OpenAI or Anthropic,
prompts and selected query context are sent to that provider.
7. Column Governance and Audit
Use Foggy MCP -> Column Policies to block or mask fields exposed to AI clients.
Use Foggy MCP -> Audit Logs to review MCP and AI Chat activity. Audit records include
user, channel, tool name, status, and error summary when available.
8. Basic Verification
- Run the Settings health check.
- Call
tools/list from an MCP client and confirm that tools are returned.
- Call
dataset.list_models and confirm that only accessible models are listed.
- Ask AI Chat a simple question if you enabled AI Chat and data transfer consent.
- Export a query result from AI Chat if Excel export is part of your workflow.
9. Troubleshooting
- If embedded mode fails to start, confirm that
asyncpg is installed in the Odoo Python environment.
- If Excel export fails, confirm that
openpyxl is installed.
- If an MCP client receives authentication errors, confirm that the bearer token starts with
fmcp_ and belongs to an active user.
- If AI Chat says LLM data transfer is not enabled, enable AI Chat LLM Data Transfer Consent in Settings.
- If provider calls fail, verify provider, model, API key, base URL, and network access from the Odoo server.
- If a user sees fewer models or fields than expected, review Odoo access rights, record rules, company access, and Foggy column policies.
10. Support
Contact support@foggysource.com with:
- addon version,
- Odoo version,
- deployment mode,
- error message or screenshot,
- reproduction steps,
- whether AI Chat uses no LLM, a self-hosted endpoint, or a cloud provider.