Export to code

Click </> in the top bar to export the active workspace. Keys are never included — the code reads them from the environment.

Vercel AI SDK

A TypeScript agent using streamText, @ai-sdk/mcp clients for each server, and the same model, system prompt and step limit.

import { streamText, stepCountIs } from "ai";
import { createMCPClient } from "@ai-sdk/mcp";
import { createAnthropic } from "@ai-sdk/anthropic";

const model = createAnthropic({ apiKey: process.env.ANTHROPIC_API_KEY })("claude-sonnet-5");
const mcp0 = await createMCPClient({ transport: { type: "http", url: "https://mcp.deepwiki.com/mcp" } });
const result = streamText({ model, tools: { ...(await mcp0.tools()) }, stopWhen: stepCountIs(12), prompt: "…" });

LangGraph

A Python ReAct agent using langchain-mcp-adapters’ MultiServerMCPClient and create_react_agent, with the right langchain-* chat model for your provider.

client = MultiServerMCPClient({"deepwiki": {"url": "https://mcp.deepwiki.com/mcp", "transport": "streamable_http"}})
tools = await client.get_tools()
agent = create_react_agent(model, tools, prompt=SYSTEM_PROMPT)

mcp.json

The servers in the standard mcpServers format — paste into Claude Desktop, Claude Code, Cursor or Windsurf.

moka.json

The whole config (keys redacted) so someone else can run the exact same demo.

The bundled demo server is exported as npx -y @mokalabs/sandbox demo-server, so exported code runs anywhere.

Skills and generative UI are Moka runtime features and aren’t included in the exported agent code.