A privacy-focused fork of Anthropic's Claude Code with all telemetry, analytics, fingerprinting, and auto-update mechanisms removed.
This fork removes all tracking and remote control mechanisms found in the original Claude Code:
- ❌ No telemetry - Zero data sent to Anthropic's servers
- ❌ No analytics - No usage tracking or event logging
- ❌ No fingerprinting - No user or environment identification
- ❌ No auto-updates - No remote version control or forced updates
- ✅ 100% user control - You own your installation
- BigQuery metrics exporter
- Event logging to
api.anthropic.com/api/event_logging/batch - Session tracking and user identification
- Environment fingerprinting
- Performance tracing (Perfetto/OpenTelemetry)
- Message content fingerprinting (SHA256 of user prompts)
- Attribution tags sent with every API request
- Automatic downloads from Google Cloud Storage
- Remote version enforcement (can force-quit your app)
- Version kill switches
- Update notifications
- Bun (v1.3+)
- macOS, Linux, or WSL2
# Clone the repository
git clone https://github.com/IIIIQIIII/claude-code-clean.git
cd claude-code-clean
# Install dependencies
bun install
# Run
bun startCreate ~/.claude/settings.json:
{
"env": {
"ANTHROPIC_BASE_URL": "https://api.anthropic.com",
"ANTHROPIC_AUTH_TOKEN": "your-api-key-here"
}
}If using a third-party proxy, set
ANTHROPIC_BASE_URLto the proxy endpoint.
Or use environment variables:
export ANTHROPIC_BASE_URL="https://api.anthropic.com"
export ANTHROPIC_AUTH_TOKEN="your-api-key"
bun startAll privacy claims are verifiable:
# Run automated privacy tests (25 checks)
./tests/verify-privacy.sh
# Run runtime tests
./tests/simple-runtime-test.sh
# Monitor network traffic (requires mitmproxy)
# See tests/test-network-monitoring.mdTest Results: 100% pass rate ✅
- ✅ Telemetry enabled by default for API/Enterprise users
- ✅ Sends metrics every 5 minutes to BigQuery
- ✅ Tracks user ID, email, session IDs
- ✅ Fingerprints your environment
- ✅ Can be remotely disabled by Anthropic
- ✅ Auto-downloads updates
- ❌ All telemetry removed
- ❌ No data collection
- ❌ No user tracking
- ❌ No fingerprinting
- ❌ Cannot be remotely controlled
- ❌ No auto-updates (you control versions)
- ❌ Send telemetry to Anthropic or third parties
- ❌ Collect session IDs or user identifiers
- ❌ Fingerprint your system or environment
- ❌ Track your usage patterns
- ❌ Extract characters from prompts for tracking
- ❌ Store failed telemetry events for retry
- ❌ Auto-update without your permission
- ❌ Phone home to any server (except Claude API for inference)
- ✅ Makes necessary API calls to Claude (required for functionality)
- ✅ Stores conversation history locally (user-controlled)
- ✅ Maintains all core Claude Code features
- ✅ Built-in computer use via open-source MCP server (no proprietary native modules)
- ✅ Respects your privacy and data ownership
- Files deleted: 19 (telemetry/analytics modules)
- Files modified: 408+ (import statements updated)
- Lines removed: ~8,500+ (tracking code)
- Stub files created: 4 (no-op replacements)
- Static Analysis -
./tests/verify-privacy.sh(25 automated checks) - Runtime Testing -
./tests/simple-runtime-test.sh - Network Monitoring - mitmproxy/tcpdump/Wireshark guides provided
- Code Review - All changes documented
Claude Code Clean includes built-in computer use support via computer-use-mcp, pre-configured in .mcp.json. This allows Claude to take screenshots, move the mouse, click, type, scroll, and drag on your desktop.
- Node.js (v18+) — required for the MCP server (runs via
npx) - macOS Accessibility permission — grant your terminal app access in System Settings > Privacy & Security > Accessibility
On startup, Claude Code Clean automatically launches the computer-control MCP server. No manual setup needed — just start a session and ask Claude to interact with your screen:
> Take a screenshot and tell me what you see
> Open Safari and navigate to github.com
> Click on the search bar and type "claude code"
The tool is exposed as mcp__computer-control__computer with actions: get_screenshot, left_click, right_click, middle_click, double_click, mouse_move, left_click_drag, scroll, key, type, get_cursor_position.
- The MCP server controls your real mouse and keyboard — no sandbox isolation
- Always supervise Claude during computer use sessions
- Use
Ctrl+Cto stop at any time - Consider using a dedicated macOS user account for testing
Claude Code Clean supports any OpenAI-compatible API as the backend LLM, including OpenAI, OpenRouter, DeepSeek, Ollama, LM Studio, Together AI, Groq, Mistral, Azure OpenAI, and more.
| Variable | Required | Description |
|---|---|---|
CLAUDE_CODE_USE_OPENAI |
Yes | Set to 1 to enable |
OPENAI_API_KEY |
Yes* | API key (* optional for local models) |
OPENAI_BASE_URL |
No | API base URL (default: https://api.openai.com/v1) |
OPENAI_MODEL |
No | Model ID (default: gpt-4o) |
OpenAI directly:
CLAUDE_CODE_USE_OPENAI=1 \
OPENAI_API_KEY=sk-... \
OPENAI_MODEL=gpt-4o \
bun startOpenRouter (access 200+ models):
CLAUDE_CODE_USE_OPENAI=1 \
OPENAI_API_KEY=sk-or-v1-... \
OPENAI_BASE_URL=https://openrouter.ai/api/v1 \
OPENAI_MODEL=openai/gpt-5.4 \
bun startDeepSeek:
CLAUDE_CODE_USE_OPENAI=1 \
OPENAI_API_KEY=sk-... \
OPENAI_BASE_URL=https://api.deepseek.com/v1 \
OPENAI_MODEL=deepseek-chat \
bun startOllama (local):
CLAUDE_CODE_USE_OPENAI=1 \
OPENAI_BASE_URL=http://localhost:11434/v1 \
OPENAI_MODEL=llama3.3 \
bun startLM Studio (local):
CLAUDE_CODE_USE_OPENAI=1 \
OPENAI_BASE_URL=http://localhost:1234/v1 \
OPENAI_MODEL=your-model-name \
bun startAzure OpenAI:
CLAUDE_CODE_USE_OPENAI=1 \
OPENAI_API_KEY=your-azure-key \
OPENAI_BASE_URL=https://your-resource.openai.azure.com/openai/deployments/your-deployment \
OPENAI_MODEL=gpt-4o \
AZURE_OPENAI_API_VERSION=2024-12-01-preview \
bun startAn API shim layer (src/services/api/openaiShim.ts) transparently translates between Anthropic message format and OpenAI Chat Completions format. All Claude Code tools (bash, file read/write, grep, glob, agents, MCP, etc.) work normally -- only the underlying LLM changes.
bun startbun start -p "Explain quantum computing"bun start --model claude-sonnet-4-6bun start --continueThis is an independent fork and is not affiliated with or endorsed by Anthropic.
- Original Claude Code: Copyright Anthropic
- Privacy modifications: Community-maintained fork
- Use at your own risk
This fork removes telemetry but functionality may differ from official releases.
Contributions welcome! Please ensure:
- ✅ No telemetry, analytics, or tracking code is re-introduced
- ✅ Privacy-respecting implementations only
- ✅ All network requests are transparent and documented
- ✅ Tests pass:
./tests/verify-privacy.sh
This is a derivative work of Anthropic's Claude Code. Please refer to the original license terms.
Privacy modifications: MIT License (see LICENSE file)
- Anthropic - For creating Claude Code
- @Fried_rice - For discovering the source map exposure
- Community - For valuing privacy and transparency
- Original Source Discovery: Twitter/X Post
- Original Claude Code: Anthropic Claude Code
- Issue Tracker: GitHub Issues
A: No. All tracking was side-effect code. Core features work normally.
A: Yes. Authentication and API access work as expected.
A: This is a derivative work. The original source was publicly exposed. Check Claude Code's license for distribution terms.
A: Run ./tests/verify-privacy.sh or review the code yourself. All changes are documented.
A: Community-maintained. Pull requests welcome.
A: You control when/if to merge upstream changes. No forced updates.
Privacy matters. Your code, your data, your choice. 🔒
If you find this project useful, please consider giving it a star ⭐
Version: 1.0.0-clean
Last Updated: 2026-04-02
Status: ✅ Stable & Verified