climux
Unified CLI multiplexer for AI coding agents, enabling orchestration of multiple coding CLI tools through a single interface with session persistence, cost tracking, and MCP tool support.
README
Climux
Unified CLI multiplexer for AI coding agents
Climux provides a single interface for AI agents to orchestrate multiple coding CLI tools (Claude Code, Codex, Gemini CLI, OpenCode). Instead of learning each tool's unique syntax, agents use one consistent API.
Why Climux?
- One Interface: Single command syntax for all providers
- Smart Routing: Automatically selects the best CLI for each task type
- Session Persistence: Resume interrupted work across restarts
- Parallel Workspaces: Isolate work with git worktrees
- Cost Tracking: Monitor token usage and API costs
- Graceful Fallback: Automatic provider switching when one fails
Installation
# Global install
npm install -g @veithly/climux
# Or run directly with npx
npx @veithly/climux run "your task"
Prerequisites: At least one CLI must be installed:
npm install -g @anthropic-ai/claude-code # Claude Code
npm install -g @openai/codex # Codex
npm install -g @google/gemini-cli # Gemini CLI
npm install -g opencode # OpenCode
Usage
Run a Task
# Auto-route to best provider
climux run "add user authentication with JWT"
# Specify provider
climux run "fix the login bug" --provider codex
# Run autonomously (non-interactive)
climux run "refactor database layer" --mode task
# Run in background
climux run "implement caching" --background
Session Management
climux session list # List all sessions
climux session list --status active # Filter by status
climux session resume <id> # Resume session
climux session show <id> # Show session details
climux session log <id> # View conversation log
climux session export <id> --format md # Export to markdown
climux session delete <id> # Delete session
Workspace Management
climux workspace info # Current workspace info
climux workspace list # List known workspaces
climux workspace switch /path/to/project # Switch workspace
climux workspace alias api ~/projects/api # Create alias
climux run "task" --workspace @api # Use alias
Git Worktree Integration
climux worktree create feature-auth # Create worktree
climux worktree create hotfix -b main # From specific branch
climux worktree list # List worktrees
climux worktree switch feature-auth # Switch to worktree
climux worktree delete feature-auth # Delete worktree
Monitoring
climux status # Real-time status
climux stats # Today's summary
climux stats --by provider # Per-provider breakdown
climux stats --from 2024-01-01 # Historical data
Configuration
climux config show # Show current config
climux config show --providers # Show provider status
climux config set defaultProvider codex # Set default
climux config init # Initialize config
Configuration File
Global: ~/.climux/config.yaml
defaultProvider: claude-code
routing:
- pattern: "frontend|react|vue|css"
provider: gemini-cli
- pattern: "debug|fix|bug|error"
provider: codex
- pattern: ".*"
provider: claude-code
fallbackOrder:
- claude-code
- codex
- gemini-cli
- opencode
concurrency:
maxActiveSessions: 5
maxSessionsPerWorkspace: 3
monitoring:
trackTokens: true
trackCost: true
Project-level: .climux/config.yaml (overrides global)
defaultProvider: gemini-cli
presets:
quick-fix:
mode: task
provider: claude-code
MCP Server
Climux can run as an MCP server, allowing other AI systems to use it as a tool:
climux mcp serve
Available MCP Tools:
| Tool | Description |
|---|---|
run_task |
Execute a coding task |
session_list |
List sessions |
session_resume |
Resume a session |
session_send |
Send message to active session |
get_status |
Get system status |
worktree_create |
Create git worktree |
worktree_list |
List worktrees |
get_session_stats |
Get statistics |
MCP Client Example:
const result = await client.callTool('run_task', {
task: 'add input validation',
mode: 'task',
workspace: '/path/to/project'
});
Skills & MCP Management
climux skill list # List available skills
climux skill sync # Sync skills across providers
climux mcp list # List installed MCPs
climux mcp install <package> # Install MCP package
For AI Agents
See SKILL.md for detailed agent instructions.
Quick Reference
# Parallel feature development
climux worktree create feature-x
climux run "implement feature X" --worktree feature-x --mode task
# Check before starting heavy work
climux status
# Monitor costs
climux stats --by provider
# Resume interrupted work
climux session list --status paused
climux session resume <id>
# Export for documentation
climux session export <id> --format md > session.md
Best Practices
- Use
--mode taskfor autonomous operations - Create worktrees for parallel development
- Check
climux statusbefore resource-intensive tasks - Use aliases for frequent projects (
@api,@web) - Monitor costs with
climux stats
Data Storage
- Config:
~/.climux/config.yaml - Database:
~/.climux/climux.db(SQLite) - Tables: sessions, session_logs, session_stats, workspace_aliases
Error Handling
| Error | Cause | Solution |
|---|---|---|
PROVIDER_NOT_AVAILABLE |
CLI not installed | Install or auto-fallback |
SESSION_NOT_FOUND |
Invalid session ID | List sessions |
PROCESS_CRASHED |
CLI crashed | Auto-resume |
RATE_LIMITED |
API limit hit | Auto-fallback |
Development
git clone https://github.com/veithly/climux.git
cd climux
pnpm install
pnpm build
pnpm test
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。