micro-cc 0.2.24

Harness that gives frontier models full system access — shell, filesystem, browser, MCP — running directly on the metal.

micro · cc — cognitive compute

  ███╗   ███╗██╗ ██████╗██████╗  ██████╗
  ████╗ ████║██║██╔════╝██╔══██╗██╔═══██╗
  ██╔████╔██║██║██║     ██████╔╝██║   ██║
  ██║╚██╔╝██║██║██║     ██╔══██╗██║   ██║
  ██║ ╚═╝ ██║██║╚██████╗██║  ██║╚██████╔╝
  ╚═╝     ╚═╝╚═╝ ╚═════╝╚═╝  ╚═╝ ╚═════╝
       c o g n i t i v e  c o m p u t e

Knowledge work is, by extension, a coding problem.

Frontier models with full system access — shell, filesystem, browser, MCP — running directly on the metal. Verbose, thorough planning paired with unrestricted file operations means micro·cc can take on complex multi-step work autonomously. Point it at any project directory and go.

Install

Please use iTerm2. Highly recommended terminal UI.

pip install micro-cc

On first launch, micro·cc creates ~/.micro-cc/ with an example .env — add your API keys there and restart. That directory is also where conversation history and project state live.

Recommended terminal: iTerm2 (macOS).

Platform

Built and tested on macOS. Works on Linux. Windows should work via pip install + any modern terminal — iTerm2 is macOS-only but micro·cc doesn't depend on it, it just looks best there.

Environment

Edit ~/.micro-cc/.env — set one endpoint:

# Option 1: Anthropic direct + OpenAI for embeddings
ANTHROPIC_API_KEY=sk-ant-
OPENAI_API_KEY=sk-proj-

# Option 2: LiteLLM proxy (Bedrock, Azure, etc.)
# LITELLM_BASE_URL=https://your-proxy.com
# LITELLM_API_KEY=sk-...

# Web search (always needed for search tools)
SERPAPI_KEY=

Usage

microcc /path/to/your/project

Controls:

• Enter — submit prompt
• Shift+Enter — newline
• Escape — interrupt
• /clear — reset conversation
• /model — switch model
• /exit — quit

Data

~/.micro-cc/
  .env                              API keys
  projects/
    {project}_{hash}/
      messages.jsonl                conversation history
      summary.json                  sliding-window summary
      project_path.txt              maps hash back to directory

Tools

Tool	Description
bash_	Shell execution in project_dir
read_	Read files with line numbers, offset/limit
write_	Create/overwrite files, auto-creates dirs
edit_	Surgical string replacement (fails if ambiguous)
glob_	Find files by pattern, sorted by mtime
grep_	Regex search with context lines
browser_	Web browsing and page extraction (beta)
computer_use_	Screen interaction and GUI automation (beta)

All tools are discoverable — use search_tools to find more at runtime.

Relative paths resolve to project_dir. Absolute paths work anywhere.

Architecture

start_live_.py (CLI)                textual TUI
       │
       │ async for event in claude_loop()
       ▼
claude_loop_.py (Core)              API calls, tool execution, JSONL storage
       │
       │ execute_tool_call()
       ▼
tools/                              bash_, file ops, browser, computer_use
       │
       ▼
~/.micro-cc/projects/{hash}/        conversation persistence

Full system access plus verbose planning. The model reads your codebase, writes plans, executes tools, observes results, iterates — until the work is done.