RAG服务器

docker compose up --build

# Windows PowerShell
$env:LLM_MODEL="your-model-identifier"; docker compose up -d

# bash / macOS / Linux
LLM_MODEL=your-model-identifier docker compose up -d

docker compose down

docker compose down; docker volume rm mcp-rag_chroma_data

docker compose exec mcp-rag pytest tests/ -v

Smaug is black with bright pink polka dots.
(source: /app/sample_docs/Later_edits.txt)

The Lord of the Eagles, named Kevin, leads the group. He possesses a commercial
pilot's license issued by the Misty Mountains Aviation Authority. As part of his
community service agreement, he is obligated to rescue at least one wizard per
fiscal year. On days off he drives a Toyota Camry and enjoys playing sudoku.
(source: /app/sample_docs/Later_edits.txt)

"Farewell, good thief. I go now to the halls of waiting to sit beside my
fathers, until the world is renewed..."
"No! There is more in you of good than you know, child of the kindly West.
Some courage and some wisdom, blended in measure. If more of us valued food
and cheer and song above hoarded gold, it would be a merrier world.
But sad or merry, I must leave it now. Farewell!"
(source: Chapter_18_THE_RETURN_JOURNEY.txt)

Chapter 4 — "Over Hill and Under Hill" covers the company's journey through
the Misty Mountains, their capture during a storm, and their descent into
the goblin tunnels.
(source: Chapter_04_OVER_HILL_AND_UNDER_HILL.txt)

# Instead of the user's question phrasing:
find_relevant_docs("travel to goblins caves")  # poor recall

# Try content-rich terms:
find_relevant_docs("goblins caves mountains captured dwarves hobbit")  # hits Ch.4

fastmcp              # MCP server framework
langchain            # LLM orchestration
langchain-community
langchain-openai     # LM Studio / OpenAI-compatible LLM client
langchain-ollama     # Ollama LLM integration (optional / bonus embeddings)
langgraph            # Corrective RAG graph engine
chromadb             # Vector store (in-process, no server)
chardet              # File encoding auto-detection
unstructured         # Document loaders
pypdf                # PDF support
python-docx          # Word document support
python-dotenv        # Environment variable management
pyyaml               # YAML config support
pytest               # Testing
pytest-asyncio
pytest-mock

winget install Python.Python.3.11

git clone <repo-url>
cd mcp-rag

# Windows
py -3.11 -m venv .venv
.venv\Scripts\Activate.ps1

# macOS / Linux
python3.11 -m venv .venv
source .venv/bin/activate

pip install -r requirements.txt

ollama pull nomic-embed-text

LLM_BASE_URL=http://localhost:11434/v1
LLM_MODEL=llama3.2
LLM_API_KEY=ollama
python -m src.server

# docker-compose.yml environment section
LLM_BASE_URL: https://api.openai.com/v1
LLM_MODEL: gpt-4o-mini
LLM_API_KEY: sk-...

# Windows PowerShell
$env:LLM_BASE_URL="https://api.openai.com/v1"
$env:LLM_MODEL="gpt-4o-mini"
$env:LLM_API_KEY="sk-..."
docker compose up -d

python -m src.server

{
  "servers": {
    "mcp-rag": {
      "type": "stdio",
      "command": "${workspaceFolder}/.venv/Scripts/python.exe",
      "args": ["-m", "src.server"],
      "cwd": "${workspaceFolder}"
    }
  }
}

{
  "servers": {
    "mcp-rag": {
      "type": "http",
      "url": "http://localhost:8000/mcp"
    }
  }
}

pytest tests/ -v

docker compose exec mcp-rag pytest tests/ -v

mcp-rag/
├── .github/
│   └── copilot-instructions.md
├── src/
│   ├── server.py        # FastMCP server & tool definitions
│   ├── config.py        # Configuration parameters
│   ├── indexer.py       # Document loading, chunking, ChromaDB
│   ├── graph.py         # LangGraph Corrective RAG pipeline
│   └── prompts.py       # Prompt templates
├── tests/
│   ├── test_indexer.py
│   ├── test_graph.py
│   └── test_mcp_tools.py
├── sample_docs/
│   └── Hobbit/          # Hobbit chapters split by chapter for better source attribution
├── results/
│   ├── ARCHITECTURE.md  # Detailed graph & design documentation
│   └── REPORT.md        # Development history & lessons learned
├── .vscode/
│   └── mcp.json         # VS Code MCP client configuration
├── .venv/               # Python 3.11 virtual environment (not committed)
├── docker-compose.yml
├── Dockerfile
└── requirements.txt