Skip to content

Connect an AI agent to My Architect with MCP

Connect an MCP-compatible agent to let it read and update the same project model you use in My Architect. The MCP package provides tools. The My Architect workflow plugin adds instructions for context reading, work ownership, coordination, and verification.

Open API Keys in the application, generate a key, and copy it when displayed. The Connect your AI agent section provides configurations for supported clients. Prefer its current account-specific instructions when they differ from an example here.

The published package, @my-architect/mcp, runs locally over stdio and forwards requests to the configured backend. It requires Node.js 20 or newer.

Setting Hosted configuration
MCP_API_KEY Your exact key from the application
MA_API_URL https://my-architect.app

Without MA_API_URL, the process defaults to http://localhost:3100. That is the local development backend, not the hosted service. A local stdio process can still use the hosted backend; it does not make a local copy of your project.

Store your real key using your host’s supported credential or environment configuration. Do not commit it to a repository or paste it into shared documentation. The examples below contain placeholders.

The application provides this MCP-only setup command:

Terminal window
codex mcp add my-architect \
--env MCP_API_KEY=mcp_YOUR_TOKEN \
--env MA_API_URL=https://my-architect.app \
-- npx -y @my-architect/mcp@latest

Check the configured server with codex mcp list. Use your host’s current plugin installation interface if you also want the workflow plugin; adding an MCP server alone does not install skills.

The plugin setup installs the MCP configuration and workflow skills. Make the key available to the shell from which you launch Claude Code:

Terminal window
export MCP_API_KEY=mcp_YOUR_TOKEN

Then use Claude Code’s plugin commands:

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

Verify the connection with /mcp. For MCP tools without the plugin, the application offers:

Terminal window
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

Claude Desktop, Cursor, and compatible JSON clients

Section titled “Claude Desktop, Cursor, and compatible JSON clients”

Add the server through the client’s supported MCP configuration. For clients accepting an mcpServers object, the configuration is:

{
"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"
}
}
}
}

Use the location and restart procedure specified by your client. For OpenClaw through mcporter, the application’s Connect section provides the corresponding command. Do not paste the JSON into a client that requires a different configuration format.

Restart the MCP process after installation or updates; an already running process does not acquire newly installed code. Also restart the relevant agent session when updating its workflow instructions.

Ask the agent to confirm the exact intended project binding and read a known project item. list_projects can help locate a project, but neither the newest result nor an empty response establishes which project you intended. Preserve spaces and punctuation in the project ID.

Read live capabilities before relying on coordination or delivery operations. Visible tool names and successful authentication do not prove that every backend operation is supported or currently permitted.

If access fails, check the backend URL, active key, client process, and returned error. Keep the last verified state; do not initialize replacement data to make a failed read appear successful.

Continue with the agent quickstart or multi-agent coordination.

Verified against app/MCP 1.11.0 and workflow plugin 2.6.0 on October 13, 2026. Confirm the connected server’s capabilities before applying a procedure.