Prompt-caching ships auto breakpoints for Claude
prompt-caching is an open-source MCP plugin that auto-injects Anthropic cache_control breakpoints and exposes cache stats for token and cost visibility. It targets repeated coding/agent workflows and claims major savings on repeated turns, especially outside Claude Code's built-in automatic caching path.
This is a practical developer-ops release: it productizes prompt-caching mechanics that many teams still implement inconsistently.
- –Automates cache breakpoint placement for system prompts, tool definitions, and repeated file/context patterns.
- –Adds observability (`get_cache_stats`, cacheability analysis) that helps teams debug misses instead of guessing.
- –Benchmarks and README position break-even around early repeated turns, with large upside on long coding sessions.
- –Key caveat: Claude Code already auto-caches its own internal calls, so strongest value is custom Anthropic SDK/app traffic.
- –MIT license and MCP-first design make it easy to trial in Cursor/Windsurf/other compatible clients.
DISCOVERED
75d ago
2026-03-13
PUBLISHED
75d ago
2026-03-13
RELEVANCE
AUTHOR
ermis