MCP server

With the Model Context Protocol (MCP), an AI assistant calls tools to answer a question. FirstSI's MCP server presents your data as tools the assistant uses on its own, for instance when asked which computer has had a network problem since this morning, which countries a server talks to, or which alerts are open.

#Prerequisites

You need an API token carrying the scopes for the data you want to open (see API tokens). A tool whose scope is missing is not even offered to the assistant. You also need an MCP-compatible assistant that supports the HTTP transport ("Streamable HTTP" or "remote MCP server", depending on the product).

#Connection settings

The MCP server is read-only. Even with a token holding write:* scopes, no MCP tool writes anything; write operations only go through the REST API.

#Declare the server in an MCP client

Most clients accept a declaration like this one (the exact field names vary from one client to another):

{
  "mcpServers": {
    "firstsi": {
      "type": "http",
      "url": "https://console.example.com/api/v1/mcp",
      "headers": { "Authorization": "Bearer ${FIRSTSI_TOKEN}" }
    }
  }
}

#Check the connection

Once connected, the assistant should list FirstSI's tools, for example netdiag_overview, list_machines or flows_destinations. Then ask it something simple, such as "List the offline computers". If the tool list is empty, the token probably has no read scope.

Without an assistant, you can ask for the tool list from the command line:

curl -s https://console.example.com/api/v1/mcp \
  -H "Authorization: Bearer $FIRSTSI_TOKEN" -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

#The tools

The full list, with each tool's arguments, is in MCP tools. They read the same data as the REST API, with the same scopes and the same rate limit.

#Privacy

The assistant only sees what the token allows, and only your customer's data. Every tool call goes through the public API: it counts against the token's rate limit and is recorded in its log (Public API screen).

On the other hand, the data that is read goes to the provider of the assistant you connect. Choose it with that in mind. FirstSI's built-in AI Assistant can run on a model hosted on your own premises.

Source: · FirstSI Docs · updated 2026-10-10