---
title: "MCP core for an LLM assistant"
url: "https://ramenm.com/ru/projects/enhanced-mcp-agent"
markdown_url: "https://ramenm.com/ru/projects/enhanced-mcp-agent/index.md"
locale: "ru"
content_language: "ru"
page_kind: "project"
source: "localized_path"
llms_url: "https://ramenm.com/llms.txt"
llms_full_url: "https://ramenm.com/llms-full.txt"
---

# MCP core for an LLM assistant

- Case study URL: https://ramenm.com/ru/projects/enhanced-mcp-agent
- Markdown URL: https://ramenm.com/ru/projects/enhanced-mcp-agent/index.md
- Role: AI Tooling Developer
- Period: 2025
- Team: Solo

Backend core для LLM assistant с plugin execution, hot reload, tool chains и explicit context handoff.

## Problem
Когда assistant вырастает за пределы demo, plugins и tool calls становится сложно развивать без поломки runtime.

## Solution
Собрал FastAPI execution core с plugin lifecycle management, hot reload, cascading tool calls и CLI client с execution history.

## Impact
Новые tools можно добавлять без переписывания assistant core, а tool execution проще inspect.

## Stack
- Python, FastAPI, MCP, Tool calling, LLM APIs, Docker / Docker Compose, TypeScript

## Metrics
- Plugin model: Hot reload
- Execution: Context handoff
- Surface: CLI + API

## Highlights
- Сделал plugin hot reload без restart core process.
- Собрал explicit context handoff между chained tools.
- Разделил runtime responsibilities и plugin responsibilities.

## Lessons
- AI tooling становится useful, когда execution behavior predictable.
- Clean plugin boundary важнее количества подключенных models.