Skip to main content
MCP server CX Codex

crit — Claude Code MCP server: features and installation

This MCP server for Claude Code exposes crit commands as typed tools. An agent can manage runs, evaluate models, and start an inference server through the connection.

Apache-2.0 ★ 50 updated 2026-11-16

Install into Claude Code · Platform CLI All methods and platforms ↓
claude mcp add crit -- uvx crit
Open on GitHub

What crit does

The server passes crit's command-line functions to an agent over MCP stdio. Tools cover run records and evaluation; long commands continue as jobs that the agent can wait on or inspect through their logs. Tool argument schemas come from crit's own CLI parser.

There are also tools for kernels, model operations, and inference methods. The agent runs crit plugins, crit smoke, crit bench, and crit hub separately in a shell. Run the MCP server on the machine hosting the engines, and use absolute paths in tool calls.

What it helps with

  • plan a crit run and manage its records
  • start an evaluation and follow its job
  • build and sweep kernel templates on a device
  • serve a candidate and inspect server status

Example prompts for the agent

  • Plan a crit run for this project.
  • Run an evaluation for the model and show its status.
  • Build a kernel from a template and check it.

How to install crit

Platform CLI

claude mcp add crit -- uvx crit

uvx runs the command named after the package; if the README names another one, use uvx --from crit <command>

Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more

Questions about crit

How do I install crit in Claude Code?

Run claude mcp add crit -- uvx crit.

What does crit do?

It exposes crit CLI commands as MCP tools so an agent can manage runs, evaluations, models, and inference tasks.

What does crit need to run?

crit itself is required. Run the MCP server on the machine where the compute engines run.

Similar MCP servers

All MCP servers for Agent setup: memory, context, orchestration →

Project sponsors

Companies that keep this analytics open