Configuration
Every setting in MCP Server for Odoo, and what each one changes.
There is almost nothing to configure. The permission model, the exposed data and the audit trail all belong to AI Assistant for Odoo; this module puts them on the wire.
Before you start
Configure exposure in the core module first. If nothing is exposed there, tools/list returns an empty list and a connected client will have nothing to do.
Issuing a token
AI Assistant > Connections > Issue a token
| Field | What it means |
|---|---|
| What is it for | A name you will recognise when revoking. "Claude Desktop on my laptop" |
| Speaks for | The person whose access rights every request will use |
| Expiry | Optional and recommended |
Press Issue. The token is shown once, with a ready-made client configuration block. Copy it now: it is stored only as a SHA-256 digest and cannot be recovered.
A user may issue tokens that speak for themselves. Only a manager may issue one that speaks for somebody else.
The client configuration
The wizard hands you exactly this, with your own URL and token filled in:
{
"mcpServers": {
"odoo": {
"url": "https://your-odoo.example.com/mcp/v1",
"headers": { "Authorization": "Bearer tvai_..." }
}
}
}Paste it into your client's MCP configuration. The commonest failure is putting the token somewhere other than the Authorization header.
Revoking
AI Assistant > Connections > Tokens, then Revoke. It takes effect on the next request. The audit rows survive revocation, because the interesting question is usually asked afterwards.
Governance
- A token's owner and secret are frozen after issue. Revoke and reissue instead - otherwise revocation would be a rename.
- Set an expiry. A token with none outlives the reason it was issued.
- The token list shows last used, request count, and the client's own name for itself, so a credential nobody remembers issuing can be found.