Roo Code

Roo Code is the open-source AI coding agent that runs as a VS Code extension. Wire it to agentry so Roo can spin up sandboxes, write code, run commands, and ship apps on your server.

You'll need

• The agentry CLI installed and signed in. See Quick start, step 1–2.
• VS Code with the Roo Code extension installed.
• ~3 minutes.

Add agentry as an MCP server

Roo Code reads MCP config from mcp_settings.json inside its extension storage. The easiest way to open it: in VS Code, open the Roo Code panel → click the ⚙ settings icon → MCP Servers → Edit MCP Settings.

Add agentry to the mcpServers block:

{
  "mcpServers": {
    "agentry": {
      "command": "agentry",
      "args": ["mcp"],
      "disabled": false,
      "alwaysAllow": []
    }
  }
}

Save the file. Roo reloads MCP servers without a restart.

Verify

In the Roo Code panel, switch to a new task and ask:

Which agentry tools do you have?

Roo should list sandbox_create, app_probe, command_run, file_write, port_wait, project_start, and the rest. If they show up, you're wired.

If they don't, click the MCP Servers tab in the Roo panel — agentry should show a green dot. A red dot means the server failed to start; click for the stderr.

Auto-approve agentry tools

Roo prompts before every tool call by default. For an agentry session you trust, add tools to alwaysAllow so the work flows without confirmation prompts:

{
  "mcpServers": {
    "agentry": {
      "command": "agentry",
      "args": ["mcp"],
      "disabled": false,
      "alwaysAllow": [
        "sandbox_create",
        "sandbox_list",
        "app_probe",
        "command_run",
        "file_write",
        "file_read",
        "file_list",
        "file_grep",
        "file_replace",
        "port_wait",
        "project_start",
        "project_list",
        "project_logs"
      ]
    }
  }
}

Skip the destructive ones (sandbox_delete) unless you really want one-click teardown.

Try it

Use agentry to build a hello-world homepage — a single page that says "Hello from agentry". Start the dev server.

Roo should call the tools in sequence and report the port the dev server is on. It won't hand you a public link — open the sandbox in the dashboard and click Share for a *.agentry.live preview URL.

Path issues

If Roo says agentry failed to start, the CLI probably isn't on its $PATH. Use the full path:

which agentry

Then in mcp_settings.json:

{
  "mcpServers": {
    "agentry": {
      "command": "/Users/you/.local/bin/agentry",
      "args": ["mcp"]
    }
  }
}

Switch between models with OpenRouter

Roo Code supports OpenAI-compatible providers natively. Open Roo's settings → API Provider → choose OpenAI Compatible, set the base URL to https://openrouter.ai/api/v1, paste your OpenRouter API key, and pick a model from the list (anthropic/claude-3.5-sonnet, openai/gpt-4o, meta-llama/llama-3.3-70b-instruct, …).

One key, every model. Swap models with a dropdown. agentry doesn't care which one drives the tool calls.

Common gotchas

Tools work but the agent never calls them

Check the Roo system prompt. If it's been heavily customized, it may not include the MCP tool-use guidance Roo ships by default. Reset to defaults and re-test.

Roo says "MCP server not connected"

• Confirm agentry mcp runs from a terminal without errors.
• Confirm you're signed in: agentry status should print your device + current server.
• If you have a corporate proxy, MCP stdio still works (it doesn't use HTTP), but agentry login and agentry mcp need to reach api.agentry.run.

Next

• Prompting — patterns Roo will follow.
• Ship an app — go from preview to production URL.