MCP Servers
Model Context Protocol (MCP) servers extend what AI models can do during a conversation by providing them with tools — capabilities such as web search, database queries, file access, or any custom action exposed by a compatible server.
MCP servers are configured in the Settings panel under the MCP Servers section. You need the MCP Management permission to make changes here.
Adding an MCP server
Click New Server to open the add server form.
Fill in the following:
- Name — A display name for this server
- Endpoint — The full URL of the MCP server's endpoint
- Bearer Token (optional) — If the server requires authentication, enter the token here
Click Save. The Gateway will immediately attempt to connect to the server and retrieve its tool list.
Connection status
After saving, each server card shows its connection status:
| Status | Meaning |
|---|---|
| Spinner | Connection test in progress |
| Connected (green) | Successfully connected and tools retrieved |
| Failed (red) | Could not connect — check the endpoint URL and any authentication requirements |
Controlling which surfaces a server is available on
Each server has two toggles that control where its tools are made available:
- Chat — Makes the server's tools available in OptimaChat conversations
- API — Makes the server's tools available to API consumers using the
/v1/chat/completionsendpoint
These can be enabled or disabled independently. For example, you might enable a server for Chat but not for API, or vice versa.
Viewing available tools
If a server is connected and exposes tools, click the tool count below the server's toggles to expand the list. Each tool shows its name and a short description.
If a connected server exposes no tools, this is indicated with "No tools exposed by this server".
Removing an MCP server
Click the delete button on the server card. The server is removed immediately and its tools will no longer be available on any surface.