← Back to Discover

cc-code

A terminal coding agent built in Rust that reads existing Claude Code configuration and works with Anthropic, OpenAI-compatible, DeepSeek and GLM models.

Coding agentsCommand-line tools
Visit website ↗

What it does

cc-code is a coding agent that runs in the terminal. It is written in Rust and, according to its README, starts quickly and uses around 50MB of memory, which it contrasts with agents built on Node.js or Bun. It can reuse an existing Claude Code setup directly, including the .claude folder, agents, skills, hooks and MCP servers, so there is no migration step. It is not tied to a single model vendor. The README lists Anthropic and OpenAI-compatible APIs plus DeepSeek and GLM, and says the system prompt is kept frozen so that prompt caching hits 95-99% of the time. Core tools are always loaded while the rest are loaded on demand through tool search. The default Auto permission mode checks each tool call in two layers: your own rules plus a built-in deny list, then a separate judge model that asks for confirmation only when it is unsure. Other capabilities include background sub-agents that run in parallel, automatic compaction of long sessions, streaming Markdown rendering and the ACP protocol for connecting editors such as Zed. It installs with npm (npm install -g @cc-claw/code) or with an install script on macOS, Linux and Windows. The page warns that on npm 12 and later the postinstall script that downloads the binary has to be allowed explicitly. The interface supports English and Simplified Chinese, and the project is released under the Apache 2.0 licence.

Capabilities

  • Rust-built terminal agent with a small memory footprint
  • Reuses Claude Code configuration, agents, skills, hooks and MCP servers
  • Works with Anthropic, OpenAI-compatible, DeepSeek and GLM models
  • Auto permission mode with a separate judge model for risky commands
  • Parallel background sub-agents and automatic compaction of long sessions
  • ACP support for connecting editors such as Zed

When to use it

  • You already use Claude Code and want a lighter agent that reads the same configuration
  • You want a terminal coding agent that can run against DeepSeek, GLM or other OpenAI-compatible models
  • You want to drive the agent from an ACP-capable editor