Your AI Studio,
scriptable by any agent.

The Model Context Protocol is an open JSON-RPC 2.0 standard for letting AI agents call tools. This project ships an MCP server as an HTTP endpoint at /api/mcp, authenticated with a bearer token. Any compatible client connects once and gets a toolbox that creates, edits, and reads files in your AI Studio workspace.

1. 1

   Authorize

   Client sends Authorization: Bearer <MCP_AUTH_TOKEN>. The server validates it timing-safe and rejects without touching JSON-RPC.

2. 2

   Initialize

   Client calls { "method": "initialize" }. Server returns protocol version and capabilities.

3. 3

   List & call

   tools/list returns all 8 tools; tools/call runs one — e.g. create_file({ path, content }).

4. 4

   Land it

   The write lands in the shared workspace store and shows up live in the Workspace Console.

Every MCP client takes a server config. For the HTTP transport, that's a url and a headers block carrying the bearer token. Your exact URL and token live on the Workspace Console — copy them from the Connect tab.

Name: AI Studio MCP
MCP server URL: https://<your-preview-url>/api/mcp

Note: Do NOT fill in the "Advanced" fields (Client ID / Client Secret).
Leaving Advanced blank connects directly with your MCP server without requiring an OAuth provider.

{
  "mcpServers": {
    "ai-studio": {
      "url": "https://<your-preview-url>/api/mcp",
      "headers": { "Authorization": "Bearer <MCP_AUTH_TOKEN>" }
    }
  }
}

curl -X POST https://<your-preview-url>/api/mcp \
  -H "Authorization: Bearer <MCP_AUTH_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

{ "jsonrpc": "2.0", "id": 2, "method": "tools/call",
  "params": {
    "name": "generate_component",
    "arguments": {
      "name": "PricingCard",
      "description": "A pricing card with tier name, price, and 3 features"
    }
  }
}