# juyterman1000/entroly [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/juyterman1000/entroly  
**GitHub Stars:** 480  
**npm Downloads (last month):** 727  
**Views:** 4  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/juyterman1000-entroly

## Description
Auditable context control plane and MCP server for AI coding agents. Compresses context 70–95% (BM25 + entropy + dep-graph knapsack), stabilizes prompt prefixes for provider cache discounts, routes easy tasks to cheaper models (RAVS Bayesian router), and verifies answers locally with WITNESS hallucination guard (0.844 AUROC, $0, 3 ms). MemoryOS adds local budget-aware working/episodic/semantic memory with decay, safety scanning, and durable persistence. 38 agent integrations (Cursor, Claude Code, Codex, Aider, and more). Ships as MCP server (entroly serve), HTTP proxy, or Python/Rust library. Apache-2.0, local-first. pip install entroly

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "entroly": {
    "command": "npx",
    "args": ["-y","entroly@latest"]
  }
}
```

## Documentation

## What juyterman1000/entroly does

juyterman1000/entroly adds a local evidence-selection layer for AI coding tools. It examines repository material, chooses content within a token budget, and records what it included, what it omitted, and the reasoning behind the selection. The result is a receipt that can be inspected later.

Omitted material remains recoverable through content-addressed references, provided the referenced store and original source bytes are still available. This makes context reduction reversible rather than a permanent deletion of repository evidence.

The project also supports durable repository evidence for continuing work between agents. The `entroly-work resume` flow can reconstruct unfinished work for a receiving agent, which must review and acknowledge the inferred state before taking ownership.

## How it works

Entroly operates locally and can be used as an MCP server, an HTTP proxy, or a Python or Rust library. With a verified prompt hook, it can run before a model plans. MCP-only usage exposes callable tools that an agent may choose not to invoke, while API traffic is intercepted only when routed through the Entroly proxy.

Context selection combines budgeted evidence handling with receipt-backed recovery. The project description identifies BM25 ranking, entropy, and dependency-graph knapsack selection as part of its context compression approach. It also describes prompt-prefix stabilization for provider caching and a Bayesian router for directing simpler work to lower-cost models.

The installation includes local checks such as `entroly verify-claims` and `entroly simulate`. These commands operate on the local project without an API key and help inspect behavior before connecting a paid model or provider.

## Setup and configuration

The recommended installation is the Python package:

```bash
pip install -U entroly
```

From the repository you want to inspect, run:

```bash
entroly verify-claims
entroly simulate
```

`simulate` accepts a task-specific query. Other supported installation paths include the npm package, a Rust source build with the proxy feature, Homebrew on macOS or Linux, and the published Docker image. Package-runner options include `uvx`, `pipx`, `npx`, `pnpm dlx`, and `bunx`.

The Python distribution is the complete option for the MCP server, CLI, SDK, proxy, verification, and native-engine paths. The Node package provides a local Node/WASM CLI, while the Rust build produces a standalone program.

## Tools and capabilities

- Select repository evidence under a token budget.
- Compress context while preserving recoverable references to omitted bytes.
- Produce receipts describing retained and excluded material.
- Recover source evidence through content-addressed storage.
- Verify claims and simulate selection locally.
- Store working, episodic, and semantic memory with budget awareness, decay, safety scanning, and durable persistence.
- Support agent continuity through repository-backed work resumption.
- Serve context functions through MCP or route API traffic through an HTTP proxy.

## Limitations and notes

Exact recovery depends on keeping the referenced store and source bytes available. Receipt-backed recovery does not guarantee answer quality. Prompt hooks require host activation, so an installed integration should not be assumed to be active; `entroly activation status --json` checks that state.

The local verification and simulation commands do not call an AI, prove activation, demonstrate answer quality, or establish a lower provider bill. Cost and cache effects depend on the workload, provider, and routing path. MCP integrations are callable tools, so an agent may skip them unless the surrounding workflow uses them.

_Full upstream README: https://allmcps.com/mcp/juyterman1000-entroly/readme_