Claude-Zeroclaw 🦞⚡

"The crab that pinches between conversations—your zero-cost cognitive claw for Claude Code orchestration."

🧠 What Is Claude-Zeroclaw?

Claude-Zeroclaw is a decentralized sidecar daemon that sits between your terminal and Claude Code CLI sessions. It acts as a contextual router, sub-agent orchestrator, and prompt augmentation layer—all without incurring additional token costs beyond your standard Claude API usage.

Think of it as a cognitive claw that reaches into your local filesystem, your ongoing conversation threads, and your MCP tool definitions, then pinches the most relevant context into every Claude Code invocation—automatically, transparently, and at zero additional overhead.

Unlike traditional proxy solutions that simply forward requests, Zeroclaw understands your workflow topology: it learns which files you edit together, which skills you invoke in sequence, and which Claude Desktop sessions you reference most often. It then builds a context graph that evolves across your entire development session.

🦀 Core Metaphor: The Hermit Crab

Imagine a hermit crab that carries its shell from place to place. Claude-Zeroclaw is that shell—but instead of a home, it carries your conversational context:

• 🐚 Session memory persists across CLI invocations
• 🦞 Sub-agent claws reach into MCP tools and SDKs
• ⚡ Zero-cost means no hidden fees, no premium tiers, no token waste

This is not a proxy. This is not a router. This is a context ecosystem.

📦 Features

🔁 Context Routing Engine

• Routes prompts through configurable skill chains before reaching Claude
• Automatically injects relevant file snippets, previous responses, and session state
• Supports dynamic skill switching based on detected intent

🧩 Sub-Agent Orchestration

• Decompose complex tasks into sub-agent workflows
• Each sub-agent runs with its own context window, skill set, and temperature
• Results are merged back into the main conversation seamlessly

🖥️ Desktop ↔ CLI Bridge

• Syncs context between Claude Desktop sessions and Claude Code CLI
• No more copy-pasting context between interfaces
• Bidirectional state transfer via local MCP server

🌐 Multilingual Prompt Augmentation

• Automatically translates non-English prompts to Claude's optimal input language
• Returns responses in your original language
• Supports 42 languages natively; extensible via custom plugins

🕐 24/7 Customer Support Integration

• Pairs with any support ticketing system via MCP tools
• Enables Claude Code to triage, respond, and escalate issues autonomously
• Full audit log of all automated support interactions

📱 Responsive UI Overlay

• Terminal-based TUI that displays current context state, active sub-agents, and token usage
• Collapsible panels for skills, files, and chat history
• Works in any terminal that supports ANSI escape codes

🖥️ OS Compatibility

🔧 Architecture Overview

graph TB
    User([User Terminal]) -->|Command| CLI[Claude Code CLI]
    CLI -->|HTTP| Zeroclaw[Claude-Zeroclaw Daemon]
    
    subgraph "Zeroclaw Engine"
        Router[Context Router]
        Skills[Skill Chain]
        SubAgents[Sub-Agent Pool]
        Memory[Session Memory]
        MCP[MCP Server Bridge]
    end
    
    Zeroclaw --> Router
    Router --> Skills
    Router --> SubAgents
    Router --> Memory
    
    Skills -->|Enriched Prompt| API[Claude API / OpenAI API]
    SubAgents -->|Parallel Tasks| API
    Memory -->|Persistence| Disk[(Local Disk)]
    
    MCP ---|Bidirectional| Desktop[Claude Desktop]
    MCP ---|Bidirectional| SDK[Claude SDK]
    MCP ---|Events| Integration[3rd Party Tools]
    
    API -->|Response| Router
    Router -->|Final Output| CLI
    CLI --> User

⚙️ Installation & Setup

Prerequisites

• Node.js 20+ or Python 3.11+
• Claude Code CLI installed and authenticated
• (Optional) Claude Desktop for desktop↔CLI bridging

Quick Install

# Via npm
npm install -g claude-zeroclaw

# Via Homebrew (macOS)
brew install claude-team/tap/zeroclaw

# Via Docker
docker pull claude-zeroclaw:latest

Verify Installation

zeroclaw --version
# Output: Claude-Zeroclaw v2.3.1 (2026)

🎛️ Example Profile Configuration

Create ~/.zeroclaw/config.yaml:

# ~/.zeroclaw/config.yaml
profile: developer-fullstack

context:
  max_session_tokens: 32000
  inject_recent_files: true
  recent_file_count: 5
  session_persistence: ~/.zeroclaw/sessions/

routing:
  skill_chain:
    - sanitize-prompt
    - augment-context
    - decompose-task
    - route-to-subagent
  fallback_strategy: aggregate

sub_agents:
  enabled: true
  max_concurrent: 3
  agent_definitions:
    - name: code-reviewer
      model: claude-4-5-sonnet
      temperature: 0.2
      skills: [linter, best-practices, security-scan]
    - name: test-writer
      model: claude-4-5-sonnet
      temperature: 0.7
      skills: [jest-config, mock-generation, coverage-analysis]
    - name: documentation
      model: gpt-4o
      temperature: 0.4
      skills: [markdown-lint, jsdoc, swagger-gen]

mcp_bridge:
  port: 4123
  sync_desktop: true
  sync_frequency_ms: 5000

multilingual:
  enabled: true
  default_input_lang: auto
  preserve_technical_terms: true

integration:
  openai_api_key: ${OPENAI_API_KEY}
  claude_api_key: ${CLAUDE_API_KEY}
  customer_support:
    provider: zendesk
    auto_triage: true
    escalation_threshold: 0.85

🚀 Example Console Invocation

Basic Command

# Start Zeroclaw daemon
zeroclaw daemon --profile developer-fullstack

# In another terminal, use Claude Code as usual
claude "Refactor the authentication middleware to use JWT"

Advanced Invocation with Sub-Agents

# Invoke Claude Code through Zeroclaw with explicit sub-agent routing
claude --zeroclaw "Deploy the microservice to staging" \
  --subagent code-reviewer \
  --subagent test-writer \
  --context-inject recent:3 \
  --session-resume last

Desktop ↔ CLI Sync

# Sync current Claude Desktop session to CLI
zeroclaw sync --from desktop --to cli --session analysis-123

# Result: Claude Code CLI now has all context from Desktop session
claude "Continue the analysis from where I left off on Desktop"

🔌 API Integration

OpenAI API + Claude API Dual Mode

Claude-Zeroclaw supports routing to either API based on task complexity:

Configure via environment variables:

export CLAUDE_API_KEY=sk-ant-...
export OPENAI_API_KEY=sk-proj-...
export ZEROCLAW_ROUTING_POLICY=cost-optimized

MCP Tool Integration

Zeroclaw exposes a local MCP server that any MCP-compatible tool can connect to:

// In your MCP client config:
{
  "mcpServers": {
    "claude-zeroclaw": {
      "command": "zeroclaw",
      "args": ["mcp-server"],
      "env": {
        "ZEROCLAW_CONFIG": "~/.zeroclaw/config.yaml"
      }
    }
  }
}

🛠️ Skill System

Zeroclaw includes a built-in skill library that can be extended:

📊 SEO-Friendly Keywords (for discoverability)

This repository actively targets the following search terms naturally throughout its documentation and metadata:

• Claude Code CLI context management
• Zero-cost AI orchestration (note: "zero-cost" refers to no additional software fees, not API costs)
• MCP tools integration for Claude
• Claude sub-agent workflow automation
• Multilingual Claude assistant
• Claude Desktop to CLI sync
• Sub-agent decomposition for Claude Code

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

⚠️ Disclaimer

Claude-Zeroclaw is an independent open-source tool and is not affiliated with, endorsed by, or sponsored by Anthropic or OpenAI. All trademarks, service marks, and company names referenced herein—including "Claude," "Claude Code," "Claude Desktop," "OpenAI," and "GPT"—are the property of their respective owners.

This software is provided "as is" without warranty of any kind, express or implied. Users are responsible for:

• Complying with the terms of service of any third-party APIs they connect
• Ensuring they have appropriate API keys and billing configured
• Safeguarding sensitive d