A governance layer that sits between Claude Desktop and any MCP server you point at. Every tool call gets evaluated against policies you define: allow, deny, or pause for human approval. When approval is required, a reviewer sees full context (agent identity, operation, parameters, risk level) in a dashboard before the action executes. Everything generates an audit trail with hash-chained traces. Useful when you want agents to use powerful tools like filesystem access, database writes, or API calls, but need a human gate on risky operations. Works with existing MCP servers without modification. Part of the broader SidClaw platform that also wraps LangChain, OpenAI, and other agent frameworks.
claude mcp add --transport stdio io.github.vladuzh-sidclaw-governance-mcp uvx sidclaw-governance-mcp