Using Claude Code CLI Bridge with OpenClaw

Option 1: Direct Shell Tool (RECOMMENDED)

OpenClaw can already call shell scripts directly. Use this in your prompts:

# Instead of relying on default model, explicitly call the bridge
exec:./scripts/claude-cli-bridge.sh "your prompt here" /tmp

Option 2: Set via Environment Variable

export OPENCLAW_MODEL="shell:/Users/andy/.openclaw/workspace/scripts/claude-cli-bridge.sh"
openclaw gateway restart

Option 3: Per-Request Model Override

In any OpenClaw session, you can override the model:

/model shell:/Users/andy/.openclaw/workspace/scripts/claude-cli-bridge.sh

Option 4: Create a Skill

Create a skill that wraps the bridge:

{
  "name": "claude-max",
  "description": "Use Claude Code CLI with Max plan",
  "command": "/Users/andy/.openclaw/workspace/scripts/claude-cli-bridge.sh"
}

Testing the Bridge

cd ~/.openclaw/workspace
./scripts/claude-cli-bridge.sh "What is 2+2?" /tmp

Notes

Slower than API: CLI has ~2-3s startup time
Uses Max plan: No per-call charges
Good for: Heavy usage sessions where startup time doesn't matter
Not ideal for: Quick queries where latency matters

Current Status

✅ Bridge script: Working (authenticated with Max plan) ✅ Direct test: claude --print responds correctly ⚠️ Full integration: Requires OpenClaw restart or skill wrapper

FilesExpand file tree

CLAUDE_CLI_BRIDGE_SETUP.md

Latest commit

History