Atla MCP Server

What is Atla MCP Server?

An MCP server implementation that provides a standardized interface for LLMs to interact with the Atla SDK, enabling access to state-of-the-art evaluation models like Selene 1. It is intended for developers building AI agents that need automated quality evaluation of responses.

How to use Atla MCP Server?

Install the server by cloning the repository, installing dependencies with uv venv and uv sync, and setting the ATLA_API_KEY environment variable. Then configure it as an MCP server in OpenAI Agents SDK, Claude Desktop, or Cursor using the provided JSON configurations.

Key features of Atla MCP Server

• Evaluate individual responses using Selene 1.
• Run batch evaluations with Selene 1.
• List available evaluation metrics.
• Create new evaluation metrics.
• Fetch metrics by name.

Use cases of Atla MCP Server

• Automatically evaluate and improve LLM responses during agent execution.
• Perform iterative refinement by evaluating responses for cliché or helpfulness.
• Integrate standardized evaluation into multi-agent workflows.

FAQ from Atla MCP Server

What is Atla MCP Server used for?

It provides a standardized interface for LLMs to evaluate responses using Atla’s evaluation models, such as Selene 1.

What dependencies are required?

You need Python, uv installed on your system, and an ATLA_API_KEY. For OpenAI Agents integration, you also need the OPENAI_API_KEY environment variable.

Where does the API key come from?

Your ATLA_API_KEY can be obtained from your Atla account at https://www.atla-ai.com/sign-in.

How does the server communicate?

The server uses the stdio transport protocol, configured as a local MCP server via command-line execution.

What integrations are supported?

Atla MCP Server can be used with OpenAI Agents SDK, Claude Desktop, and Cursor MCP clients.