About pash-mcp-server
pash-mcp-server is a standalone MCP server extracted from the Felix ecosystem, intended for MCP clients that need tools loaded from a local directory. It uses PASH compression to reduce large JSON responses and registers tools dynamically from manifests.
Key features
- 01Dynamic tool loading from manifests
- 02PASH compression for large JSON responses
- 03stdio transport
- 04Configuration through .env
- 05Standalone implementation without Felix core dependencies
Use cases
- 01Expose locally configured MCP tools to a client
- 02Reduce token use for large tool responses
- 03Load tools from an mcp_engines directory
MCP support
It acts as an MCP server over stdio, loading tools from `mcp_engines` manifests and routing requests to the configured MCP servers. PASH compression is applied to large JSON responses.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
We haven’t collected a package or remote endpoint for this server. Follow the instructions in the project README.
Tools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project READMEFelix Standalone MCP Server
Standalone MCP Server with PASH Compression (Phoenix Architecture)
Этот проект представляет собой минимальный, production-ready Model Context Protocol (MCP) сервер, извлеченный из экосистемы Felix. Он разработан для предоставления внешних инструментов (Cursor, Claude Desktop и др.) с максимальной экономией токенов благодаря встроенному PASH-сжатию.
Особенности (Phoenix Architecture)
- Чистое ядро: Никаких скрытых зависимостей от монолитной кодовой базы Феликса.
- Динамическая регистрация: Автоматическое сканирование директории
mcp_enginesи загрузка инструментов изmanifest.json. - PASH Compression: Интеллектуальное сжатие тяжелых JSON-ответов (экономия >80% токенов) с прозрачным возвратом хэша.
- Zero-config запуск: Работает через
stdioс конфигурацией через.env.
⚠️ Безопасность и Runtime
Для разработчиков, вносящих изменения в серверную часть (sse_server.py, тесты): строго соблюдайте паттерн Surgical Process Cleanup by Port. Глобальное убийство процессов (taskkill /IM) категорически запрещено, так как оно обрывает агентскую сессию.
Architecture Boundaries
Felix MCP Standalone is a self-contained product. It shares principles, not code with the internal Felix core (C:\Users\53\Felix).
- Product Judge (in
src/product_judge.py) is an independent implementation of deterministic validation. It shares the principle of "structural validation before output" with the internal Mirror Forge Synthesis, but has zero shared code by design. This is intentional technical debt prevention. - Cognitive Loop and PASH compression are implemented directly in this repository. They do not import from or depend on the internal Felix core. This ensures the product can be deployed, tested, and maintained independently.
Why this matters: When you install this package, you get a complete, autonomous system. No hidden dependencies on internal infrastructure.
Установка
git clone https://github.com/AkynZH/pash-mcp-server.git
cd pash-mcp-server
pip install -e .
Конфигурация
Скопируйте .env.example в .env и настройте пути:
MCP_ENGINES_DIR=./mcp_engines
LOG_LEVEL=INFO
PASH_MIN_THRESHOLD=500
Создайте директорию mcp_engines и поместите в неё подпапки с manifest.json и исполняемыми файлами ваших MCP-серверов.
Запуск
python server.py
Сервер запустится в режиме stdio и будет готов к подключению через совместимые клиенты (например, Claude Desktop).
Структура проекта
pash-mcp-server/
├── server.py # Точка входа (FastMCP)
├── src/
│ ├── config.py # Pydantic-конфигурация
│ ├── router.py # Сканер и JSON-RPC прокси
│ └── compressor.py # Ядро PASH-сжатия
├── config/
│ └── manifest.example.json # Пример манифеста
└── tests/ # Изолированные тесты
Odysseus Integration Status
This repository serves as the cognitive core designed to be paired with an interface like Odysseus.
- ✅ Completed & Shipped: Cognitive Loop, PASH compression, Product Judge, and heuristic routing are fully implemented and tested.
- 🗺️ Roadmap: Native, one-click UI integration with the Odysseus agent (combining local reflexes with this cognitive core) is currently in active development.
We believe in shipping what works and clearly labeling what is being built.
Лицензия
MIT License (c) 2026 AkynZH
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.