# Integrations

> Source: https://whisperer.one/llms/integrations.md · Whisperer machine-readable layer.
> Product overview: https://whisperer.one/llms/product.md · Full corpus: https://whisperer.one/llms-full.txt

What Whisperer connects to, and the shape of each connection.

## Calling services: none, deliberately

Zoom, Google Meet, Microsoft Teams, Telegram, Discord and anything else work without an
integration. The desktop app captures system audio and the screen locally and sits on top of
whatever is running. There is no bot to invite, no plugin to install, no meeting link to hand
over, and consequently nothing that breaks when a calling service changes its API.

Reference: /docs/en/integrations/supported-services.html

## Calendars: two-way sync

Google Calendar, Outlook and CalDAV. A Whisperer video room can live inside an event, and
guests join by link.

## Connected services

Notion, Slack, HubSpot and Zapier connect from the Whisperer account. Once connected, Leo can
act in them on the user's behalf inside Whisperer.

These tools are **never** projected to external MCP clients — not sending mail, not posting to
chat, not writing to a CRM, and not even reading a connected drive. The user granted that
access to Whisperer; handing it to an arbitrary external client would pass someone else's
consent down the chain, which is not what was agreed to and not visible to them afterwards.

## MCP, both directions

Whisperer is an MCP **server**: external AI clients read and act on Whisperer data with the
user's consent (`/llms/mcp.md`).

Whisperer is also an MCP **client**: the user can attach their own MCP servers so Leo can use
them. Servers that run over STDIO are launched by the desktop client on the user's own
machine, never on Whisperer infrastructure — running arbitrary user-supplied commands on
shared infrastructure would be handing every user code execution on our servers.

## Export

Markdown, SRT, VTT and JSON for meeting material.