How to Use Model Context Protocol (MCP)

What is MCP?

MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external tools and services. With it, an assistant can do more than chat—read and write files, drive apps, query data—depending on which MCP servers you attach.

Why it matters

Most AI assistants are limited to the conversation itself. If you want them to:

Edit documents
Drive apps to complete workflows
Fetch weather or query a database
Control other desktop applications (such as Mind Elixir)

you need a bridge. MCP is that bridge: a common protocol for exposing tools to AI, with operations running safely on your device, and the freedom to swap servers and models as needed.

How MCP works

Take Mind Elixir as an example. You want the AI to add a branch to a mind map, but Claude or ChatGPT has no built-in knowledge of Mind Elixir's internal canvas structure. The flow looks like this:

You say: "Add a branch to the mind map"
The AI understands the intent and sends a standardized MCP request
Mind Elixir's MCP server receives it and creates the branch on the active canvas
The result goes back over MCP, and the new branch appears in real time

Three core pieces:

AI assistant — parses natural language and decides which tool to call
MCP protocol — defines how both sides communicate
MCP-enabled tools (e.g. Mind Elixir) — perform the actual actions

Clients that support MCP

Claude Desktop

Anthropic's official desktop client for Claude
One of the first clients with MCP support; works seamlessly with the @mind-elixir/mcp bridge package

Cursor

Popular AI code editor with native support for both Command (stdio) and HTTP MCP servers

Claude Code

Anthropic's developer CLI environment
Highly flexible config; supports stdio and HTTP transports for scripting and automation

Cline / Windsurf / VS Code

Common editor extensions with custom MCP server support

More clients: https://modelcontextprotocol.io/clients

Using MCP with Mind Elixir

Mind Elixir Desktop ships with a built-in MCP server. Once connected, you can inspect, generate, and edit mind maps using plain language.

What you can do

Inspect map structure: Let the AI read the full hierarchy and topics of the active mind map
Smart editing: Add, update, or remove nodes in natural language (e.g. "Add three sub-branches under Architecture")
Generate full mind maps: Transform articles, meeting notes, or project specs into complete mind maps
Summaries & arrows: Automatically create summary nodes or draw relationship arrows between topics

How to connect

Step 1: Open Mind Elixir App

Launch Mind Elixir Desktop
Create or open the mind map you want to work on (MCP operates on the active map)

Open Mind Elixir App

Step 2: Add the MCP server in your AI client

Mind Elixir supports two connection methods. Choose the one that best fits your client:

Method 1 (Recommended): Via `@mind-elixir/mcp` Command Bridge

For Claude Desktop, Cursor, and other stdio-based clients, we provide the official NPM bridge package @mind-elixir/mcp. No manual network handling required—it connects instantly.

Requirement: Node.js installed locally (version >= 18).

1. Claude Desktop Setup

Open your Claude Desktop config file (typically ~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%\Claude\claude_desktop_config.json on Windows) and add the following under mcpServers:

{
  "mcpServers": {
    "mind-elixir": {
      "command": "npx",
      "args": ["-y", "@mind-elixir/mcp@latest"]
    }
  }
}

Save the file and restart Claude Desktop.

2. Cursor Setup

In Cursor, go to Settings → MCP, then click Add new MCP server:

Name: mind-elixir
Type: command
Command: npx -y @mind-elixir/mcp@latest

3. Claude Code Setup

Run the following command directly in your terminal:

claude mcp add mind-elixir -- npx -y @mind-elixir/mcp@latest

Method 2: Direct HTTP (Streamable HTTP)

The latest MCP specification has transitioned to Streamable HTTP (the legacy SSE transport has been deprecated). Mind Elixir listens on local port 6595, with the MCP endpoint at /mcp.

If your client supports connecting directly to HTTP MCP endpoints:

Server URL: http://127.0.0.1:6595/mcp

Client commands & configs:

Claude Code:

claude mcp add --transport http mind-elixir http://127.0.0.1:6595/mcp

Codex:

codex mcp add --url http://127.0.0.1:6595/mcp mind-elixir

Generic JSON Configuration (for HTTP-capable clients):

{
  "mcpServers": {
    "mind-elixir": {
      "url": "http://127.0.0.1:6595/mcp"
    }
  }
}

Step 3: Start using it

Once connected, make sure the target mind map is open in Mind Elixir Desktop, then ask your AI assistant:

"Read the structure of the current mind map"
"Add three branches under the root node: Ideas, Planning, Execution"
"Based on this requirement doc, generate a structured mind map"
"Analyze this mind map and suggest improvements or missing topics"

Tips

Keep it running: Mind Elixir Desktop must be open and active while you use MCP
Be specific: Clear instructions (naming specific parent nodes or topic titles) yield the best results
Iterate: For large modifications, let the AI work in progressive steps
Save your work: Remember to review and save the mind map after edits

Troubleshooting

Check app status: Make sure Mind Elixir Desktop is running and an active mind map is open
Check port availability: Verify that port 6595 is not occupied by another program
Verify Node.js: If using npx, run npx --version to ensure Node.js is installed
Firewall / Antivirus: Ensure local loopback connections to 127.0.0.1:6595 are not blocked
Restart: Completely quit and reopen Mind Elixir Desktop, then reconnect