This solves the problem of LLMs confidently citing outdated information by applying temporal decay scoring to retrieved context. It wraps 21 adapters around sources like GitHub, Hacker News, arXiv, Reddit, SEC filings, GDELT news, and Singapore government procurement data. Every response gets tagged with publication and retrieval timestamps, then re-ranked using exponential decay functions tuned per source type. You can run it locally with full Playwright scraping or hit the hosted Cloudflare Worker endpoint. The real product is the decay-adjusted relevancy layer and the FreshContext envelope spec, not the adapters themselves. Useful when you need Claude to distinguish between a 2022 blog post and current documentation, or when building agents that need provenance and confidence scores on every piece of context.
claude mcp add --transport stdio io.github.princegabriel-lgtm-freshcontext uvx freshcontext