> ## Documentation Index
> Fetch the complete documentation index at: https://platform.atlan.com/llms.txt
> Use this file to discover all available pages before exploring further.

# OpenAI Agents SDK

> Expose Atlan MCP tools to an OpenAI Agents SDK agent.

Connect an SDK agent to Atlan over Streamable HTTP so it can discover and use
the tools available to the authenticated identity.

<Note>
  **Preview.** This guide covers the Python Agents SDK and the current
  `MCPServerStreamableHttp` transport.
</Note>

## Prerequisites

* A Python project using a current OpenAI Agents SDK release.
* The Atlan MCP endpoint: [https://agentgateway.atlan.engineering/mcp](https://agentgateway.atlan.engineering/mcp).
* OAuth support in the runtime.
* A resolved and committed dependency lockfile for the application.

## Setup

1. Configure an `MCPServerStreamableHttp` connection for the Atlan MCP
   endpoint.
2. Configure the runtime's OAuth sign-in flow for the Atlan connection.
3. Load the resulting MCP server into the agent's `mcp_servers` collection.
4. Start with a read-only tool before enabling change-capable tools.

Do not place credentials in the endpoint URL or application source.

## Expected result

The SDK initializes the Atlan connection, loads the available tool list, and
makes those tools available to the agent for the run.

## Verification

Start with a read-only registry query. Confirm that the run completes without
an MCP connection or authentication error, then inspect the SDK trace to see
whether the model selected an Atlan tool.

## Limitations

* This integration is in Preview and is not enabled for every Atlan account.
* Tool availability follows the authenticated Atlan identity.
* The model can choose not to call a tool unless your agent policy requires it.
* Tool-list caching can hide permission changes until the connection restarts.
* Apply your organization's data-handling policy before sending content to a
  model provider.

See the [OpenAI Agents SDK MCP guide](https://openai.github.io/openai-agents-python/mcp/)
for connection lifecycle, authentication, filtering, and approval policies.