Generative UI and MCP servers (could ChatOutputRenderer be used ?)

[jjkavalam]

Can ChatOutputRenderer be used to render responses from an MCP server ?

I have an MCP server that provides tools.

These tools return structured data.

I would like to develop a VSCode extension that can render the data in an interactive way in the copilot chat box.

---

I saw that vscode.chat.registerChatOutputRenderer API has been introduced. However, looking at the example provided:

https://github.com/microsoft/vscode-extension-samples/blob/986bcc700dee6cc4d1e6d4961a316eead110fb21/chat-output-renderer-sample/src/extension.ts#L262-L265

function writeMermaidToolOutput(sourceCode: string): vscode.LanguageModelToolResult {
	// Expose the source code as a tool result for the LM
	const result = new vscode.LanguageModelToolResult([
		new vscode.LanguageModelTextPart(sourceCode)
	]);

	// And store custom data in the tool result details to indicate that a custom renderer should be used for it.
	// In this case we just store the source code as binary data.

	// Add cast to use proposed API
	(result as vscode.ExtendedLanguageModelToolResult2).toolResultDetails2 = {
		mime,
		value: new TextEncoder().encode(sourceCode),
	};

	return result;
}

it appears only LanguageModelToolResult currently support custom rendering.

Will it be possible to bind custom rendering to the result of a tool call to an MCP server ?

---

The full solution am looking to enable here is:

• A VS Code extension offers nice/interactive UI components and a mapping of those components to mime types (e.g. application/vnd.common-ui-components.table)

• MCP servers can hint that the response of a certain tool be treated as one such mime type

This could enable generative UIs more broadly in the context of MCP servers