Python SDK

Install the Python client in the trusted runtime that owns your agent or backend:

pip install authlane

Initialize the client

import os

from authlane import Authlane

with Authlane(
    api_key=os.environ["AUTHLANE_API_KEY"],
    base_url="https://app.authlane.io",
) as authlane:
    result = authlane.services.list()

if result.error:
    print(result.error.message)
else:
    print(result.data)

Every public SDK call returns a Result with data and error. Expected API failures do not raise exceptions, so your application can handle the same stable error envelope in TypeScript and Python.

Connect an external user

session = authlane.connect_sessions.create(
    external_user_id="user_123",
    allowed_services=[],
    allowed_origin="https://app.example.com",
)

allowed_services=[] snapshots every service currently enabled for the tenant. Pass explicit service IDs to limit the session. Duplicates are accepted and deduplicated by the server.

Load user-scoped tools

from authlane.adapters import generic

with Authlane(api_key=os.environ["AUTHLANE_API_KEY"]) as authlane:
    result = authlane.user("user_123").tools.list(adapter=generic())

if result.error:
    raise RuntimeError(result.error.message)

tools = result.data

The adapter executes in your runtime. It requests access-only material when a selected tool runs, then calls the provider directly. Tool inputs and provider responses never pass through Authlane.

Async applications

from authlane import AsyncAuthlane
from authlane.adapters import langchain

async with AsyncAuthlane(api_key=os.environ["AUTHLANE_API_KEY"]) as authlane:
    result = await authlane.user("user_123").tools.list(adapter=langchain())

See Framework adapters for Agno, LangChain, OpenAI Agents, Vercel AI, Mastra, and local MCP examples.