Once.
CLAUDE CODE / MCP / EXECUTION SAFETY

Add Once execution safety to Claude Code through MCP.

Once gives Claude Code access to tools for identifying consequential writes, explaining retry risk, planning protection, applying approved protection and verifying a configured Once connection.

Short answer: Install the Once Claude Code plugin from the Once marketplace. The plugin exposes @once-agent/mcp automatically so coding-agent workflows can evaluate and protect external writes that may otherwise be repeated after an ambiguous outcome.

Why this matters

A coding agent can create or modify software that performs payments, refunds, bookings, orders, provisioning, deployments, messages and other externally visible writes.

If one of those operations succeeds but its response is lost, a blind retry can perform the same real-world effect again.

Install the Claude Code plugin

Add the Once marketplace, then install the plugin:

claude plugin marketplace add stringsofthemind-oss/once
claude plugin install once@once-agent

The plugin exposes the public @once-agent/mcp server to Claude Code.

Once is also registered in the MCP Registry as io.github.stringsofthemind-oss/once.

Manual MCP configuration

If you prefer direct MCP configuration instead of the plugin, run:

npx -y @once-agent/mcp

The Once repository includes ready-to-copy Claude Code MCP configuration examples for Windows and macOS/Linux.

View the manual Claude Code + Once configuration examples on GitHub

What the Once MCP server exposes

Claim boundary: Once does not claim universal exactly-once execution. Safety depends on stable operation identity, durable Once state, the provider integration and sufficiently authoritative provider truth. When an outcome cannot be established safely, Once can preserve uncertainty instead of blindly repeating the action.

Install Once for Claude Code

claude plugin marketplace add stringsofthemind-oss/once
claude plugin install once@once-agent

Try the Once tester · Plugin source · MCP package