Glue Claude back into OpenClaw.

Uses the official Claude CLI and scrubs out Anthropic's detection triggers from the system prompt due to Anthropic not allowing its use. Tested with Telegram. As far as I can tell all functions work such as heartbeats.

X post

Requires OpenClaw 2026.4.10+, Claude Code logged in with Max, and Node.js 22+. Non-destructive, won't touch your existing config or sessions.

npm install @zeulewan/glueclaw-provider && bash node_modules/@zeulewan/glueclaw-provider/install.sh

git clone https://github.com/zeulewan/glueclaw.git && cd glueclaw && bash install.sh

See installation docs for uninstall and details.

Uses the official Claude CLI:

claude --dangerously-skip-permissions -p \
    --output-format stream-json \
    --verbose --include-partial-messages \
    --system-prompt <scrubbed prompt> \
    --model <model> \
    --resume <session-id> \
    "<user message>"

The only way this breaks is if Anthropic changes how --system-prompt or --output-format stream-json work, which would affect all Claude Code integrations.

Switch in TUI: /model glueclaw/glueclaw-opus

• Tested with Telegram and OpenClaw TUI
• Switching between GlueClaw and other backends (e.g. Codex) works seamlessly via /model
• The installer patches one file in OpenClaw's dist to expose the MCP loopback token to plugins. A .glueclaw-bak backup is created.

Uses only official, documented Claude Code CLI flags. No reverse engineering, no credential extraction, no API spoofing. Use at your own risk. Not affiliated with Anthropic. Compatible with OpenClaw's plugin allowlist system.

https://zeulewan.github.io/glueclaw/