Safe-by-Default MCP Servers for Infrastructure Governance

Description and use

A suite of MIT-licensed TypeScript MCP servers that enforce safe-by-default governance through access modes, allowlists, protected resources, destructive gating, typed confirmations, secret redaction, and dry-run capabilities.

Problem

Giving AI agents access to production infrastructure risks accidental destruction (e.g., dropping databases) due to ambiguous prompts, while restricting dangerous tools reduces utility when they are legitimately needed.

Outcome

A suite of MIT-licensed TypeScript MCP servers that enforce safe-by-default governance through access modes, allowlists, protected resources, destructive gating, typed confirmations, secret redaction, and dry-run capabilities.

Workflow

• Install an MCP server package (e.g., @dockndevai/mcp-kubernetes) via npx.
• Configure the server with environment variables for credentials and access mode (e.g., K8S_MODE=read-only).
• Add the server to an AI client (e.g., Claude Code) using the provided command.
• The server registers only tools permitted by the current access mode (read-only, read-write, or admin).
• For destructive operations, the agent must provide an explicit ALLOW_DELETE flag and a typed confirmation echoing the target name.
• The server redacts secrets before sending data to the model and emits audit logs to stderr.

Evidence

Every server shares the same layered policy engine: Access modes, Allowlists, Protected resources, Destructive gating, Typed confirmation, Secret redaction, Dry-run + audit.

Source

npx -y @dockndevai/mcp-kubernetes

Source