Your own tools and MCP

If your system isn't in the list, add it in a chatbot's Tools.

An HTTP tool

New tool describes one call to your API:

Field
Name and function name	What you call it, and what the model calls (lowercase letters, digits and `_`).
What it does	The model reads this to decide when to use the tool. Be specific: what it returns and when to use it.
Parameters	What the chatbot fills in from the conversation, e.g. `order_id`, each with a description and whether it's required. Use `{name}` in the address and `{{name}}` in headers, query and body.
Connection	Adds the API key or token. Secrets are encrypted and never shown again.
Headers, query, body	The request itself. Leave the body empty for a webhook to send the parameters as they are.
Output mapping	Only these fields reach the model: fewer tokens, nothing private by accident. Empty: the whole response.
Permission	Read, Write, Delete, Financial, Sensitive or Admin.
When it runs	Automatically, the visitor confirms, a team member approves, or never. Empty: by permission.
Signed-in customers only	Offered only to visitors your website signed in; use `{{user.id}}` for their ID. See Visitor identity.
Timeout and retries	How long to wait, and how often to try again.

Run test sends a real request and shows the request, the response and what the model would see. Calls go only to public addresses, and only to your allowed outgoing domains if you set them.

Connections

A connection stores credentials for your system (API key, token, custom header, user name and password) once, for several tools. They're encrypted, never shown again and never sent to the model.

MCP servers

Connect an MCP server, then choose which of its tools the chatbot may use and Import them. They follow the same permissions, run settings and policies as every other tool.

Webhooks

To have ORYVOX tell your system about events instead, see Webhooks.