# Framework adapters

Choose a shipped adapter after binding one authenticated external user.

Adapters translate canonical definitions into local framework tools. They do not create an
Authlane gateway.

## Prerequisites

Authenticate the SaaS user on the server, initialize the TypeScript or Python SDK, and install the
framework's optional peer.

## Implement the workflow

| Runtime | Framework | Adapter | Complete guide |
| --- | --- | --- | --- |
| TypeScript | Vercel AI SDK | `vercelAI()` | [Vercel AI](/docs/sdk/vercel-ai) |
| TypeScript | OpenAI Agents | `openAIAgents()` | [OpenAI Agents](/docs/sdk/openai-agents) |
| TypeScript | Mastra | `mastraAI()` | [Mastra](/docs/sdk/mastra) |
| TypeScript | caller-owned MCP | `mcpServer()` | [Local MCP](/docs/sdk/local-mcp) |
| Python | Agno | `agno()` | [Agno](/docs/sdk/agno) |
| Python | LangChain | `langchain()` | [LangChain](/docs/sdk/langchain) |
| Python | OpenAI Agents | `openai_agents()` | [OpenAI Agents](/docs/sdk/openai-agents) |
| Python | portable local tools | `generic()` | [Python SDK](/docs/sdk/python) |

## Python

The Python factories share one Result-handling pattern:

```python
import os

from authlane import Authlane
from authlane.adapters import agno, generic, langchain, openai_agents

with Authlane(api_key=os.environ["AUTHLANE_API_KEY"]) as authlane:
    user = authlane.user("user_123")

    agno_result = user.tools.list(adapter=agno())
    if agno_result.error is not None:
        raise RuntimeError(agno_result.error.code)
    assert agno_result.data is not None
    agno_tools = agno_result.data

    langchain_result = user.tools.list(adapter=langchain())
    if langchain_result.error is not None:
        raise RuntimeError(langchain_result.error.code)
    assert langchain_result.data is not None
    langchain_tools = langchain_result.data

    openai_result = user.tools.list(adapter=openai_agents())
    if openai_result.error is not None:
        raise RuntimeError(openai_result.error.code)
    assert openai_result.data is not None
    openai_tools = openai_result.data

    portable_result = user.tools.list(adapter=generic())
    if portable_result.error is not None:
        raise RuntimeError(portable_result.error.code)
    assert portable_result.data is not None
    portable_tools = portable_result.data
```

## Expected result

The returned object is native to the selected framework and bound to one external user.

## Handle errors

Handle the Authlane tuple or Python Result before calling the framework. Framework lifecycle and
model calls may still throw according to their own contracts.

## Security boundary

Never cache or serialize executable toolsets. Each invocation gets a fresh access-only lease and
calls the provider from the SaaS runtime; Authlane receives neither tool input nor provider output.

## Next step

Open the complete guide for your selected ecosystem.