Last updated: March 2026

Privacy Policy

prompt-caching runs entirely on your machine. No data leaves. No analytics on your prompts. No telemetry.

🔍What this plugin processes

The plugin reads the messages, system prompts, and tool definitions you pass to its MCP tools in order to inject cache_control markers. It also accumulates token usage counters in memory to show you session savings.

  • Message arrays passed to optimize_messages
  • System prompts and tool definitions you include
  • Token counts returned by the Anthropic API (input, output, cache reads)

🚫What this plugin does NOT do

  • Does not transmit any of your data to external servers
  • Does not store anything to disk — session statistics live in memory only and are lost on restart
  • Does not log, track, or analyze your prompts, code, or conversations
  • Does not include analytics, telemetry, or crash reporting of any kind
  • Does not read your files, environment variables, or system configuration beyond what you explicitly pass to a tool

🔗Third-party services

This plugin itself has no third-party integrations. When you use it as part of an AI workflow, your messages are sent to the services your client is configured to use:

  • Anthropic API — your messages are processed by Anthropic's infrastructure. See Anthropic's Privacy Policy.
  • Your AI client (Claude Code, Cursor, Windsurf, etc.) — each has its own privacy policy.

🏠Data residency

No data from this plugin leaves the machine it runs on. All processing is local and in-memory. When you stop the MCP server, all session data is gone.


📬Contact

Questions about this privacy policy? Open an issue on GitHub: