MCP isn't automatically detected in VS Code

[markgoho]

Current Behavior

Nx extension is running in VS Code, no popup to automatically start.

Expected Behavior

Starting VS Code, shortly after a popup appears to recommend starting the Nx MCP server.

Steps to Reproduce

Use VS Code Nightly which has MCP functionality, and have Nx extension running.
Compared to Cursor which does present the popup, VS Code does not.

Failure Logs / Images / Videos

Environment

• Nx Console version: 20.5.0
• VSCode version: 1.99-insider

[MaxKless]

For now, we have our custom vscode implemention of LLM tooling. This adds the @nx chat participant and for now, it's better DX than just an MCP because it provides more features and is better integrated.

You can use the @nx chat participant out of the box or add the MCP server yourself if you want.

[markgoho]

You can use the @nx chat participant out of the box

This only works in Edit/Ask mode, not Agent mode....I think the MCP will probably be better for Agent mode anyway?

or add the MCP server yourself if you want

Do you have an example of this somewhere? e.g. here's a .vscode/mcp.json file:

{
  "servers": {
    "figma": {
      "type": "sse",
      "url": "http://localhost:3333/sse"
    }
  }
}

I'm guessing npx nx-mcp ./ would start the MCP, but when I do that the process just hangs with no output, same with npx nx mcp /

[MaxKless]

Sure.... but we implement tools ourselves in VSCode already, the same as MCP tools. You can use them on insiders by building from source with the vscode.proposed.languageModelToolsForAgent.d.ts proposed API.
Track support for MCP here: microsoft/vscode#242864

If we end up deciding that MCP in vscode is the best idea, we'll implement it :) If you have any insights from your early research please share!

[markgoho]

I'm sorry Max, I'm not totally following your comment, it sounds like Nx is experimenting with other ways of exposing Nx tools to LLMs in the IDE?

If that's the case, maybe sharing the use case would be better:

In VS Code Insiders, Copilot Chat Agent mode, I want to ask the model to generate a new angular application using a workspace generator of my own creation. The result of this interaction would be a new Angular application using my workspace-plugin angular-app generator.

This is how it looks today:

My hope would be in the future the LLM could do this more intelligently, perhaps via MCP Server, perhaps through another technique you're exploring.