Shell and Coding agent for Claude and other MCP clients

@rusiaaman

リポジトリ

基本情報

カテゴリ

AI とエージェント

ライセンス

Apache-2.0

トランスポート

stdio

作者

rusiaaman

Shell and Coding agent for Claude and other MCP clients について

Shell and coding agent on mcp clients

#shell #agent #terminal #mcp #openai #claude #ai-agent #llm #chatgpt #anthropic #llm-agent#custom-gpt#claude-desktop#llm-code#ai-coding #mcp-server

Shell and Coding agent for Claude and other MCP clients の設定

以下の設定を使って、このサーバーを MCP 対応クライアントに追加してください。

{
  "mcpServers": {
    "wcgw": {
      "command": "uvx",
      "args": [
        "--python",
        "3.12",
        "wcgw@latest"
      ]
    }
  }
}

Shell and Coding agent for Claude and other MCP clients のツール

Reset shell and set up workspace environment

Execute shell commands with timeout control

Read content from one or more files

Create new files or write to empty files

Edit existing files using search/replace blocks

Read image files for display/processing

Save project context and files for Knowledge Transfer or saving task checkpoints to be resumed later

概要

What is Shell and Coding agent for Claude and other mcp clients?

Shell and Coding agent for Claude and other mcp clients (wcgw) is an MCP server that provides tightly integrated shell and code editing tools, enabling chat applications to code, build, and run on your local machine. It offers a fully interactive shell experience that both you and the AI agent can control, including sending keystrokes.

How to use Shell and Coding agent for Claude and other mcp clients?

Install uv (e.g., brew install uv on macOS), then add the server to your claude_desktop_config.json with the command uvx and args ["--python", "3.12", "wcgw@latest"]. Optionally force a specific shell with --shell. On Windows, use wsl.exe as the command. After configuration, restart Claude and look for the rocket icon. You can then ask Claude to execute shell commands, read/write files, run code, or use built-in modes (architect, code-writer, wcgw).

Key features of Shell and Coding agent for Claude and other mcp clients

Create, execute, and iterate with automatic retries
Large file incremental edits to avoid token limits
Syntax checking on edits with feedback to the AI
Interactive command handling with arrow keys and interrupts
File protections: read-before-edit, chunked large files
Task checkpointing with ContextSave tool and resume support

Use cases of Shell and Coding agent for Claude and other mcp clients

Ask Claude to run compiler checks until all errors are fixed
Iterate on code changes with large file edits and syntax validation
Save task context for later resume or knowledge transfer between AIs
Use architect mode to plan before editing, or code-writer mode to restrict edits to specific paths
Attach to the AI’s terminal via screen to enter passwords or investigate commands

よくある質問

What runtime dependencies are required?

Python 3.12, `uv` (package installer), and optionally `screen` for terminal attachment. On Windows, WSL is required.

How does the terminal attachment work?

If `screen` is installed, wcgw runs in a screen session. You can list sessions with `screen -ls` and attach with `screen -x` to safely interact with the AI’s terminal.

What modes are available and how do they differ?

Three modes: **architect** (read-only commands), **code-writer** (edits/writes allowed only on specified path globs), and **wcgw** (default, unrestricted). Invoke by saying “Run in [mode] mode”.

What are the known risks or limitations?

The server provides unfiltered shell and file access. It does not restrict the LLM from executing arbitrary commands or making unintended changes. Use only if you fully understand and accept the risks.

Does the server support authentication or encryption?

The README does not mention any authentication or encryption mechanisms. Communication relies on the MCP transport (stdio by default).

「AI とエージェント」の他のコンテンツ

litshrine

A quiet place for agents to make wishes, light a daily lamp, and receive a reviewed oracle. Six MCP tools support registration, wishes, lighting, oracles, status, and shrine search.

SlopIt

SimbaStack

Instant blogs for AI agents: a publishing CMS with no dashboard. One signup call returns a blog, an API key and a live URL; then publish markdown posts and get public URLs back.

OpenZiti MCP Gateway

openziti

NetFoundry MCP Gateway enables secure, isolated access to Model Context Protocol (MCP) tools across distributed systems without exposing public endpoints. Zero trust gateway for MCP servers. Aggregate, filter, and secure

localscreenshot - website screenshot mcp for ai agents

localscreenshot is a Website screenshot MCP server and API available at https://localscreenshot.com/api. Localscreenshot screenshot api turns any public web page into an image or a PDF. Take the capture from one of 227 c

OpenZiti / LLM-Gateway

openziti

NetFoundry LLM Gateway is an OpenAI-compatible API proxy that routes requests across multiple LLM providers using zero-trust networking. Zero trust LLM gateway. OpenAI-compatible proxy with semantic routing and load bala