Pydantic AI

OpenAIProvider takes base_url. Which model class you pick decides whether you land on chat completions or responses.

OpenAIProvider takes base_url and api_key, and is handed to a model class as provider=.

pip install pydantic-ai

from pydantic_ai import Agent
from pydantic_ai.models.openai import OpenAIChatModel
from pydantic_ai.providers.openai import OpenAIProvider

model = OpenAIChatModel(
    'zen5',
    provider=OpenAIProvider(base_url='https://api.hanzo.ai/v1', api_key='sk-...'),
)

agent = Agent(model)
print(agent.run_sync('Hello').output)

Lands on POST /v1/chat/completions.

Pick the model class deliberately, because the two land on different addresses. OpenAIChatModel, prefix openai-chat:, is what backs every OpenAI-compatible provider and reaches POST /v1/chat/completions. OpenAIResponsesModel, the bare openai: prefix and the modern default, reaches POST /v1/responses. Both are served; the chat class is the safer default against a compatible endpoint.

OPENAI_BASE_URL and OPENAI_API_KEY are the environment equivalents, and OpenAIProvider also accepts a pre-built client as openai_client.

MCP

MCPToolset takes the server URL as its first positional argument. A plain URL is treated as streamable HTTP; SSE only if the path ends in /sse.

from pydantic_ai import Agent
from pydantic_ai.mcp import MCPToolset

toolset = MCPToolset('https://api.hanzo.ai/v1/mcp')
agent = Agent(model, toolsets=[toolset])

It also reads a JSON config file carrying an mcpServers object.

MCP

On this page