Semantic Kernel

The parameter differs by language, and a single answer would be wrong for one of them.

.NET

AddOpenAIChatCompletion takes endpoint, a Uri. Custom endpoints on the OpenAI connector are experimental, so the SKEXP0010 warning has to be suppressed or the build fails.

using Microsoft.SemanticKernel;

#pragma warning disable SKEXP0010

IKernelBuilder builder = Kernel.CreateBuilder();
builder.AddOpenAIChatCompletion(
    modelId: "zen5",
    apiKey: "sk-...",
    endpoint: new Uri("https://api.hanzo.ai/v1")
);

Kernel kernel = builder.Build();

The same overload exists on builder.Services for dependency injection, and as new OpenAIChatCompletionService(..., endpoint: new Uri(...)).

Python

OpenAIChatCompletion has no base-URL parameter. Its arguments are ai_model_id, service_id, api_key, org_id, default_headers, async_client, env_file_path, env_file_encoding and instruction_role. The way in is async_client — build the OpenAI client yourself and hand it over.

from openai import AsyncOpenAI
from semantic_kernel import Kernel
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion

kernel = Kernel()
kernel.add_service(OpenAIChatCompletion(
    ai_model_id="zen5",
    async_client=AsyncOpenAI(base_url="https://api.hanzo.ai/v1", api_key="sk-..."),
))

Both land on POST /v1/chat/completions.

MCP

Python ships MCPStreamableHttpPlugin in semantic_kernel.connectors.mcp. The URL is the url parameter, and the plugin is an async context manager.

from semantic_kernel.connectors.mcp import MCPStreamableHttpPlugin

async with MCPStreamableHttpPlugin(
    name="hanzo",
    description="Hanzo cloud tools",
    url="https://api.hanzo.ai/v1/mcp",
    headers={"Authorization": "Bearer sk-..."},
) as plugin:
    kernel.add_plugin(plugin)

MCPStdioPlugin and MCPSsePlugin are the siblings, on the same url parameter. Outside a context manager, call await plugin.connect() yourself.