Connecting an AI agent (MCP)

What sets My Architect apart is the MCP server running alongside the app. Through it an AI agent (Claude or any MCP-compatible one) reads the project model and edits it directly — creating nodes, requirements, diagrams and docs, with no copy-paste back and forth.

Step 1. Get a key

Open the API Keys page in the app and click Generate new key. The key is shown once — copy it right away. The same page has a Connect block with a ready-made configuration for your agent.

Step 2. Give the key to the agent

Set the environment variable in your MCP agent's configuration:

MCP_API_KEY=<your-key>

The key authenticates the MCP tools without a browser session, so the agent works with your project on your behalf.

Quickstart config

The MCP server ships as a published npm package, @my-architect/mcp. Every client below launches it the same way — as a local process over stdio — so it works with npx and needs no extra install step. There is no remote-URL setup: the local process forwards your calls to My Architect over HTTPS using the key.

Two environment variables control it:

• MCP_API_KEY — the key from the API Keys page. Required for the hosted product; without it, calls are not authenticated. Keys look like mcp_... (yours may differ — copy the exact value from the app).
• MA_API_URL — the backend URL. Set it to https://my-architect.app for the hosted product. If unset, it defaults to http://localhost:3100 (local dev only).

Requires Node.js 20 or newer.

Copy the block for your client and replace mcp_YOUR_TOKEN with the key you copied. If anything below doesn't match what you see, the Connect block on the API Keys page always shows the exact command for your account.

Claude Code

The recommended path installs both the MCP server and the myarchitect skill via the plugin marketplace. Export your key first (the plugin reads it from the shell — it carries no token itself), then run the slash commands:

export MCP_API_KEY=mcp_YOUR_TOKEN

/plugin marketplace add d7561985/my-architect-marketplace
/plugin install my-architect@my-architect-marketplace

Verify with /mcp.

Prefer the MCP server without the skill? Add it manually instead:

claude mcp add my-architect --scope user \
  --env MCP_API_KEY=mcp_YOUR_TOKEN \
  --env MA_API_URL=https://my-architect.app \
  -- npx -y @my-architect/mcp@latest

Verify with /mcp.

Claude Desktop

Add this entry to claude_desktop_config.json:

{
  "mcpServers": {
    "my-architect": {
      "command": "npx",
      "args": ["-y", "@my-architect/mcp@latest"],
      "env": {
        "MCP_API_KEY": "mcp_YOUR_TOKEN",
        "MA_API_URL": "https://my-architect.app"
      }
    }
  }
}

Cursor

Same block, in ~/.cursor/mcp.json (global) or .cursor/mcp.json (per project):

{
  "mcpServers": {
    "my-architect": {
      "command": "npx",
      "args": ["-y", "@my-architect/mcp@latest"],
      "env": {
        "MCP_API_KEY": "mcp_YOUR_TOKEN",
        "MA_API_URL": "https://my-architect.app"
      }
    }
  }
}

Any other MCP client

Clients that read the standard mcpServers JSON use the identical block above. For clients configured through mcporter:

mcporter config add my-architect "npx -y @my-architect/mcp@latest" \
  --env MCP_API_KEY=mcp_YOUR_TOKEN \
  --env MA_API_URL=https://my-architect.app

The same key works everywhere — plugin, CLI, or raw JSON.

Step 3. What the agent can do

Through the MCP tools the agent can:

• read the project context and the current hierarchy;
• build or extend the tree (epics, features, tasks);
• add requirements and source-of-truth documents;
• create and update diagrams (WBS, USM, Event Storming, box);
• plan a release and track task statuses.

So the agent works against the same model you see — one version of the truth for humans and agents alike.

Not sure where to start? Go back to getting started or the model concepts.