Connect an MCP client

Pass a Scalekit-generated MCP URL to any MCP-compatible agent, framework, or desktop client. No additional auth configuration required.

The MCP URL you generated is a standard Streamable HTTP MCP endpoint. Any spec-compliant MCP client can connect to it. No additional auth configuration, no SDK calls, no tool schema definitions are required. The client discovers available tools automatically.

LangChain / LangGraph

1
import asyncio
2
from langgraph.prebuilt import create_react_agent
3
from langchain_mcp_adapters.client import MultiServerMCPClient
4

5
async def run_agent(mcp_url: str):
6
    client = MultiServerMCPClient(
7
        {
8
            "scalekit": {
9
                "transport": "streamable_http",
10
                "url": mcp_url,
11
            },
12
        }
13
    )
14
    tools = await client.get_tools()
15
    agent = create_react_agent("openai:gpt-4.1", tools)
16
    response = await agent.ainvoke({
17
        "messages": "Get my latest email and create a calendar reminder in the next 15 minutes."
18
    })
19
    print(response)
20

21
asyncio.run(run_agent(mcp_url))

Install dependencies:

1
pip install langgraph>=0.6.5 langchain-mcp-adapters>=0.1.9 openai>=1.53.0

Claude Desktop

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

1
{
2
  "mcpServers": {
3
    "scalekit": {
4
      "transport": "streamable-http",
5
      "url": "https://your-mcp-url-here"
6
    }
7
  }
8
}

Restart Claude Desktop. The tools appear automatically in the tool menu.

MCP Inspector

Paste the URL directly into MCP Inspector to explore available tools and make live test calls before wiring up a full agent.

Any other MCP client

The URL works with any client that implements the MCP specification using Streamable HTTP transport. Pass the URL, select the transport type, and connect. No credentials to configure.

For full end-to-end agent examples with LangChain, Google ADK, and other frameworks, see the Examples section.

Full working code is on GitHub.