AgentWall
OpenClaw and Claude said "always allow."
AgentWall blocked it anyway.
npx @agentwall/agentwall setup
Works with
OpenClaw
Claude Desktop
Cursor
Windsurf
Claude Code
Any MCP client
View on GitHub →
Read the docs →
- Blocks dangerous tool calls before they execute. Credentials, shell commands, database drops.
- YAML policy engine with deny, allow, ask. Glob matching and SQL content rules.
- Browser approval UI at localhost:7823. Works in GUI clients with no terminal.
- Independent audit log. Ground truth record of every tool call.
- Hot-reload. Edit policy.yaml and changes apply instantly, no restart needed.
- Fully reversible. agentwall undo restores all original configs in one command.
- Works with OpenClaw via native plugin intercepting every tool call before it runs.