Labsco
MCP SERVER

Scores an LLM response against criteria you write and returns a critique explaining the score — on one criterion or several at once.

Summary
An evaluation loop that no longer has a backend.

The shape is worth reading: a score plus a written critique, per criterion, callable mid-conversation so a model can grade its own draft. The Atla API behind it was shut down — leaving this as a reference rather than something to install.

What it is

An MCP server over the Atla API for LLM-as-judge evaluation. You supply a prompt, a response and the criteria; an Atla evaluation model returns a score and written feedback on the response.

What you get
  • One response scored against one evaluation criterion, with a textual critique alongside the score
  • The same response scored across multiple criteria at once, one score and critique per criterion
Requirements

The repository is archived and the Atla API is no longer active, so the tools have nothing left to call. It ran as ATLA_API_KEY=<your-key> uvx atla-mcp-server.

Setup effort

One command plus a key — ATLA_API_KEY=<your-api-key> uvx atla-mcp-server, then supply credentials