FreshContext

I asked Claude to help me find a job. It gave me a list of openings. I applied to three of them. Two didn't exist anymore. One had been closed for two years.

Claude had no idea. It presented everything with the same confidence.

That's the problem freshcontext fixes.

This repository is the integrated FreshContext MCP/Core package. Core is the context-integrity layer that scores, ranks, explains, and wraps retrieved context before it reaches an LLM or agent; MCP is the primary reference/interface implementation.

Live demo: freshcontext-mcp.gimmanuel73.workers.dev/demo — same model, same query, two completely different answers. Only the temporal layer changed.

The problem

Large language models retrieve web data semantically. Cosine similarity finds the documents that match a query best — but cosine doesn't know when a document was written.

So a 2022 blog post and a 2026 paper can score nearly identically. The model gets a context window full of stale documents and faithfully summarizes 2022 advice for a 2026 question.

That's not hallucination. That's correct summarization of corrupted retrieval.

Most RAG pipelines rank context correctly semantically but incorrectly temporally.

The layer

FreshContext is a temporal correction layer for retrieval systems. One math correction applied before context reaches the LLM:

R_t = R_0 · e^(−λt)

• R_0 — base semantic relevancy (whatever your retriever already gives you)
• λ — source-specific decay constant (HN ≈14h half-life, blogs ≈29d, academic papers ≈1.6y)
• t — hours elapsed since publication
• R_t — decay-adjusted relevancy at query time

That's the whole fix. No model swap. No re-embedding. No re-indexing. The layer drops onto whatever retrieval pipeline you already have.

The layer is the product. The 21 tools shipped with this repo are reference adapters demonstrating compatibility — useful, but commodity. The DAR engine, the freshness envelope, and the FreshContext Specification are the moat.

The standard

Every FreshContext-compatible response wraps content in a structured envelope:

[FRESHCONTEXT]
Source: https://github.com/owner/repo
Published: 2024-11-03
Retrieved: 2026-03-05T09:19:00Z
Confidence: high
---
... content ...
[/FRESHCONTEXT]

When it was retrieved. Where it came from. How confident we are the date is accurate.

The FreshContext Specification v1.2 is published as an open standard under MIT licence. Any tool, agent, or system that wraps retrieved data in this envelope is FreshContext-compatible. → Read the spec · Read the methodology

Architecture boundary

FreshContext Core is the reusable center of the current integrated package. It owns freshness scoring, envelope formatting, failure guards, shared types, rank/explain primitives, and the context-conditioned utility primitive.

MCP is the primary reference/interface implementation over Core. Claude Desktop is supported, but not required. The 21 MCP tools in this repo are reference adapters and a live interface for using the system.

The production Cloudflare Worker now uses Core-backed envelope generation. Worker-specific concerns remain outside Core: MCP transport, runtime guards, KV cache policy, cache metadata injection, JSON parse/replace cache helpers, D1 feeds, cron, rate limiting, and Store/feed scoring/provenance.

The intelligence feed

Beyond the per-call envelope, the production FreshContext deployment exposes a continuous, decay-scored, deduplicated

... View full README on GitHub