Skip to main content
There are three good starting points. Pick the one that matches your goal.

​
1. Just want to try it

Open chat.yambr.com

Hosted Open WebUI with Computer Use already installed and wired up. Sign in, pick a model, ask the assistant to “create a PowerPoint about Q1 sales” — you’ll see skills, the live browser, and file previews on a real example.
Good for: one-off experiments, demos, seeing what the system can do before committing.

​
2. Drive it from your own app (Yambr as MCP provider)

Yambr publishes only the Computer Use MCP endpoint — tools, sandboxes, file hosting. You bring your own LLM provider and wire Yambr in as one more tool server.
There is no public chat/completions endpoint on api.yambr.com. LLM requests go to your provider; your Yambr key unlocks the MCP tools.
1

Get an API key

Sign in at app.yambr.com with GitHub or Google. New accounts get approved (usually quick — ping us on Telegram if needed); default budget is $10 / 30 days. Copy the key value when it’s created — you see it once. See API keys.
2

Register the MCP endpoint in your client

The endpoint is https://api.yambr.com/mcp/computer_use, auth is Authorization: Bearer <yambr-key>. Point any MCP-capable client at it — Claude Desktop, OpenAI Agents SDK, LangChain, Cursor, n8n, self-hosted LiteLLM, Open WebUI.Example with the OpenAI Agents SDK (your OpenAI key drives the model; Yambr provides the tools):
import os
from agents import Agent, Runner
from agents.mcp import MCPServerStreamableHttp

yambr_mcp = MCPServerStreamableHttp(
    params={
        "url": "https://api.yambr.com/mcp/computer_use",
        "headers": {"Authorization": f"Bearer {os.environ['YAMBR_API_KEY']}"},
    },
    name="yambr-computer-use",
)

agent = Agent(
    name="computer-user",
    model="gpt-4o",                        # ← YOUR model, YOUR provider
    mcp_servers=[yambr_mcp],
)

result = await Runner.run(agent, "Build me a landing page for a coffee shop")
print(result.final_output)
Full walkthrough: LiteLLM gateway.
3

Preview artifacts on cu.yambr.com

Files the model creates (PDFs, screenshots, HTML) are served from https://cu.yambr.com/files/{chat_id}/... and can be embedded as iframes. Artifact URLs are public (URL is the token) and scoped to the chat id. See cu.yambr.com.

​
3. Self-host everything

Run the whole stack on your own machine. You’ll need Docker. 
git clone https://github.com/Yambr/open-computer-use.git
cd open-computer-use
cp .env.example .env
# Edit .env — at minimum set OPENAI_API_KEY (any OpenAI-compatible provider)

# 1. Start the Computer Use Server (builds the workspace image on first run, ~15 min)
docker compose up --build

# 2. In another terminal, start Open WebUI
docker compose -f docker-compose.webui.yml up --build
Open http://localhost:3000. After adding a model in Open WebUI, flip Function Calling = Native and Stream Chat Response = On — without those two, tools don’t fire. Details: Model settings. Full install guide: Self-hosting · Configuration · Docker details.

​
What next?

How it's built

Computer Use Server, sandbox container per chat, six redundant MCP-native channels for the system prompt.

Skills catalogue

docx, xlsx, pptx, pdf, playwright-cli, sub-agent, frontend-design, and more.

MCP API reference

Initialize, list tools, call tools, browse resources — plain JSON-RPC over Streamable HTTP.

Integrations

Open WebUI, Claude Desktop, LiteLLM, n8n, custom clients.