Atla MCP Server implements a standardized MCP interface that allows LLMs to query and receive evaluation feedback from the Atla API for state-of-the-art LLMJ evaluation. It provides two primary tools: evaluate_llm_response, which evaluates a single LLM response against a defined evaluation criterion and returns a dictionary with a numeric score and a textual critique; and evaluate_llm_response_on_multiple_criteria, which evaluates responses against multiple criteria and returns a list of score/critique dictionaries. Both tools rely on the Atla evaluation model to produce scoring and actionable feedback. The server is designed to be run in a Python environment with an Atla API key, and can be launched manually using uvx (via uvx atla-mcp-server). The README also documents client integration patterns for OpenAI Agents SDK, Claude Desktop, and Cursor, including configuration snippets showing how to start the MCP server and pass the ATLA_API_KEY environment variable. Note: the repository is archived, and the Atla API is no longer active.