Live tools let the assistant look something up at the moment it answers. Nothing is indexed in advance: the visitor asks, the assistant calls your system, and the answer comes from what your system returns.
Use it for anything that's different per visitor or changes constantly: order status, stock levels, appointment availability, account details, delivery tracking.
Set it up under Knowledge → MCP.
What can I connect?
Two kinds of endpoint, and the type is detected automatically when you Sync:
- A remote MCP server (Streamable HTTP).
- An OpenAPI / Swagger endpoint — just paste the spec URL, e.g.
https://api.example.com/openapi.json.
How do I connect one?
- Knowledge → MCP → Add MCP Server.
- Name it, paste the Server URL (must start with
http://orhttps://). - Add Auth headers if the endpoint needs them — one per line, as
Header-Name: value. For exampleAuthorization: Bearer abc123orX-Api-Key: …. - Save, then click Sync to load its tools.
- Click Tools and tick the ones the chatbot is allowed to call. Nothing is callable until you allowlist it.
Test probes the endpoint without calling a tool — use it to confirm connectivity and auth before you debug anything else.
Which tools should I allowlist?
Only what you'd be comfortable with a stranger triggering, because in effect a stranger can. Two categories:
- Read-only tools run automatically while the assistant drafts a reply, including in autopilot. Look-ups belong here.
- Action tools that write data are held: the visitor is asked to confirm before the tool runs. Nothing is written on the AI's say-so alone.
For a write tool you enable, you can untick Ask the visitor to confirm before this runs to let it fire without a prompt. Reserve that for low-risk actions — a waived tool runs automatically, including on autopilot. See Actions and webhooks for the same idea applied to purpose-built actions.
How do I see what the assistant actually called?
View diagnostics on the MCP tab lists recent live tool calls — time, source, tool, host, status, and latency, most recent last. This is the first place to look when an answer that depends on a tool comes back wrong or slow.
When should I use this instead of an API Source?
Live tools cost a real call on every question that needs them, and a slow endpoint makes for a slow answer. So:
- If every visitor gets the same answer and it changes slowly — a catalogue, a location list — use an API Source. It's polled and indexed, and it's free per question.
- If the answer depends on who's asking or on right now — use a live tool.