Integrations (MCP)

Out of the box the model has a handful of built-in tools. Integrations add more. Catalyst connects to external tools using MCP (the Model Context Protocol) — an open standard for exposing tools to AI. Connect an MCP server and its tools become actions the model can call, both in agentic chat and in workflow tool nodes.

Connecting a server

1. Open MCP Servers from the sidebar and click Add Server.

2. Give it a name, choose the server type (for example, a Streamable HTTP server), and enter the server’s URL.

   

3. If the server needs sign-in, open Advanced — Authentication. Catalyst supports a simple token, or a full secure sign-in flow where the server walks you through granting access — no secrets to copy and paste by hand.

4. Save. Catalyst connects, discovers the server’s tools, and they become available to the model.

Using the tools

Once a server is connected, its tools show up wherever the model can act:

• In chat, with agentic mode on, the model calls them as needed and streams the calls and results into the conversation.
• In workflows, a Tool node calls a specific tool with arguments you wire from other nodes.

If a server’s sign-in expires, Catalyst flags it inline and lets you reconnect — a run won’t silently use a stale connection.

Tip

MCP is how you let the model touch your systems — an internal API, a knowledge base, a service you run. Anything exposed as an MCP server becomes a tool the model can use, without custom code on your side.

Connecting to private systems

If the tool or service lives on a private network or your own machine, connect that machine first with a Network device so Catalyst can reach it.