KoliBri MCP Server

📡 Streaming MCP Endpoint

This server exposes the Model Context Protocol directly over HTTP and streams responses so that AI copilots see updates as soon as they are generated.

					🆕 Stream-first transport:
					
						Connect with POST /mcp and keep the request open. The
						StreamableHTTPServerTransport sends newline-delimited JSON frames, allowing Claude Desktop, Copilot Chat, and
						every MCP client to process tool output incrementally without polling.

Each request follows the JSON-RPC messages defined by the MCP specification. Submit a tools/list call to discover all available tools, then issue tools/call messages on the same HTTP stream to execute them.

Use your preferred MCP client to explore the toolbox visually. Applications such as Claude Desktop or the VS Code MCP extension render the tool catalog, argument schemas, and streaming responses automatically once they connect to /mcp.

Want to try it from your own tooling? Point any MCP-aware client directly to the hosted endpoint below.

https://example.com/mcp

Note: The indices are generated during the build process. Manual refresh is not available in deployments.

🚀 Quick Start – Zero Installation

Use the hosted KoliBri MCP Server directly without any installation. Just add the configuration to your MCP client:

🔧 VS Code Copilot (Recommended)

Create an mcp.json in your workspace root or ~/.config/mcp/mcp.json:

{ "servers": { "kolibri": { "url": "<origin>/mcp", "type": "http" } }, "inputs": [] }

🤖 Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "kolibri": {
      "url": "<origin>/mcp",
      "type": "http"
    }
  }
}

						✅ Benefits of Remote Server:
						No installation required
Always up-to-date with latest samples
Works from any device
Zero maintenance

					

💻 Alternative: Local Installation

For offline use or local development, install and run locally:

npm install -g @public-ui/mcp

Then run with kolibri-mcp or configure with stdio transport:

{ "servers": { "kolibri": { "command": "kolibri-mcp" } }, "inputs": [] }

Detailed VS Code Setup Steps

Update VS Code and the GitHub Copilot Chat extension to the latest version (MCP support currently ships in the Insider/Preview channels).
Open Settings and enable GitHub > Copilot Chat: Allow MCP so Copilot may load external servers.
Create or edit mcp.json (workspace root or ~/.config/mcp/mcp.json) with the remote configuration shown above. Replace <origin> with the deployed server URL.
Reload VS Code. In the Copilot Chat panel choose Connections → kolibri to establish the streaming session.
Ask Copilot things like @kolibri search button or @kolibri fetch sample/button/basic to receive streamed answers.

In GitHub Copilot Chat you can now write:

@kolibri show me a button sample
@kolibri how do I implement a KoliBri table?
@kolibri create an accessible form

						💡 Tip:
						
							The MCP Server gives you access to all
							136 KoliBri component samples,
							50 component specifications,
							15 guided scenarios, and
							21 documentation files directly in VS Code!
						
					