MCP Server Setup

Pragma can connect to Model Context Protocol servers and expose their tools to the AI chat. Manage servers in Settings > MCP.

Add a server

1. Open Settings > MCP.
2. Select Add Server.
3. Fill in the fields:

Field	Description
Name	Display name, for example filesystem
Command	Executable to run, for example npx or npm
Arguments	One argument per line, or whitespace separated
Environment	KEY=VALUE pairs, one per line. Lines starting with # are ignored
Autostart	Start the server automatically when Pragma launches

4. Save the server.

Each entry shows a status indicator (stopped, starting, running or error), the command line and the tool names the server exposes. Use the row actions to start, stop, restart, view logs, edit or delete a server.

Example

An MCP filesystem server that can access a project directory:

{
  "servers": [
    {
      "id": "example-filesystem",
      "name": "filesystem",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/home/user/projects"],
      "env": {},
      "autostart": true
    }
  ]
}

Configuration file

Servers are persisted to mcp.json in Pragma's application config directory, which is derived from the bundle identifier dev.pragma.ide:

Platform	Path
Linux	~/.config/dev.pragma.ide/mcp.json
macOS	~/Library/Application Support/dev.pragma.ide/mcp.json
Windows	%APPDATA%\dev.pragma.ide\mcp.json

The file is a JSON object with a single servers array. Each server has id, name, command, optional args and env, and an autostart flag. Pragma generates the id when a server is added and rewrites the file whenever servers change. Running state is session-only and is not persisted.

Editing mcp.json outside Pragma is possible; the settings panel reloads the file when the window regains focus.

Using MCP tools

Once a server is running, its tools become available to the AI chat. Server logs and tool results are visible in the MCP settings panel, which makes it easier to diagnose a server that exits immediately or fails its handshake.