An agent is only as useful as the systems it can reach. Xagent connects to them in two ways.
Native connectors
For the tools most teams already use, Xagent has built-in connectors:
- Google: Gmail, Calendar, Drive and Maps
- Microsoft: Outlook, Teams and OneDrive
- Meta: Facebook and Instagram
You authorise the account once, and agents can read and act through it within the permissions you gave. Telegram is also supported as a channel for talking to agents.
MCP for everything else
MCP, the Model Context Protocol, is an open standard for exposing tools to AI models. Your CRM, a ticketing system, an internal database or a reporting API can sit behind an MCP server. Once you add that server to Xagent, its tools show up next to the native ones, and agents use them the same way.
This is how most companies connect the systems that matter most to them, because those are usually the ones nobody else has a connector for: the internal pricing tool, the warehouse database, the old system finance still runs on.
A practical order to do it in
Start with one system and one task. A good first pairing is a read-only data source and a reporting agent, for example “every Friday, pull last week’s numbers from the warehouse and write the ops summary”. Reading is easier to check than writing.
Once that works, add tools that change things, such as updating a CRM record or creating a ticket. Watch the first few runs in the trace before you put them on a schedule.
Taking agents back into your own products
The connection also runs the other way. Agents you build can be called from your own software through the API, embedded as a chat widget, or used from code through the SDK.
The rule we’d suggest is simple. Connect a system when there’s a specific task that needs it, not because it’s on a list.
