The Problem: AI Without Context

User Question: "How do I reset my password?"

1. Question → Embedding (vector)
2. Vector Store → Similarity search
3. Find relevant chunks: "Password Reset Guide"
4. Chunks → Context window
5. LLM generates answer using context

User Request: "Search my docs for API info"

1. Agent receives request
2. MCP Client discovers available tools
3. Agent calls: search_docs("API")
4. MCP Server executes tool
5. Returns results to agent
6. Agent processes and responds

"""MCP Server with RAG-capable document search."""

import json
from pathlib import Path
from mcp.server import MCPServer

mcp = MCPServer("Knowledge Base Server")

# ── Configuration ──────────────────────────────────────────
DOCS_DIR = Path("./knowledge_base").resolve()
DOCS_DIR.mkdir(exist_ok=True)


# ── Simple in-memory vector store (for demo) ──────────────
documents = []
vectors = []

def simple_embed(text: str) -> list[float]:
    """Simple keyword-based embedding (use real model in production)."""
    words = text.lower().split()
    # Create a simple frequency vector
    vocab = ["api", "password", "reset", "login", "error", "database",
             "security", "deploy", "test", "install"]
    return [words.count(w) for w in vocab]


def add_document(path: str, content: str):
    """Add a document to the vector store."""
    doc = {"path": path, "content": content, "chunks": []}
    # Split into chunks
    paragraphs = content.split("\n\n")
    for i, para in enumerate(paragraphs):
        if para.strip():
            chunk = {"text": para.strip(), "index": i}
            doc["chunks"].append(chunk)
            vectors.append({
                "doc_path": path,
                "chunk_index": i,
                "vector": simple_embed(para),
                "text": para.strip()
            })
    documents.append(doc)


# ── MCP Tools ──────────────────────────────────────────────

@mcp.tool()
def search_knowledge(query: str, top_k: int = 3) -> str:
    """
    Search the knowledge base using semantic similarity.

    Args:
        query: Search query
        top_k: Number of results to return

    Returns:
        Relevant document chunks
    """
    if not vectors:
        return "Knowledge base is empty. Add documents first."

    query_vector = simple_embed(query)

    # Calculate similarity (cosine-like)
    scored = []
    for v in vectors:
        score = sum(a * b for a, b in zip(query_vector, v["vector"]))
        total = sum(x * x for x in v["vector"]) ** 0.5
        query_total = sum(x * x for x in query_vector) ** 0.5
        if total > 0 and query_total > 0:
            score = score / (total * query_total)
        scored.append((score, v))

    # Sort by score
    scored.sort(key=lambda x: x[0], reverse=True)

    # Return top results
    results = []
    for score, v in scored[:top_k]:
        results.append(f"[Score: {score:.2f}] {v['doc_path']}:\n{v['text'][:200]}")

    return "\n\n".join(results) if results else "No relevant documents found."


@mcp.tool()
def add_to_knowledge_base(file_path: str, content: str) -> str:
    """
    Add a document to the knowledge base.

    Args:
        file_path: Path/name for the document
        content: Document content

    Returns:
        Confirmation message
    """
    add_document(file_path, content)
    return f"Added '{file_path}' with {content.count(chr(10)) + 1} lines"


@mcp.tool()
def list_documents() -> str:
    """List all documents in the knowledge base."""
    if not documents:
        return "No documents in knowledge base"
    return "\n".join(f"• {doc['path']}" for doc in documents)


@mcp.resource("knowledge://stats")
def get_stats() -> str:
    """Get knowledge base statistics."""
    total_chunks = sum(len(doc["chunks"]) for doc in documents)
    return json.dumps({
        "documents": len(documents),
        "total_chunks": total_chunks,
        "vectors_indexed": len(vectors)
    })


# ── Initialize with sample data ────────────────────────────

sample_docs = {
    "api-guide.md": """# API Guide

Our REST API uses standard HTTP methods.
Authentication uses API keys in the Authorization header.
Rate limit: 100 requests per minute.
Base URL: https://api.example.com/v1""",

    "password-reset.md": """# Password Reset Guide

To reset your password:
1. Go to /forgot-password
2. Enter your email address
3. Check your inbox for reset link
4. Click the link and create new password
5. Password must be 8+ characters""",

    "deployment.md": """# Deployment Guide

Deploy to production:
1. Run tests: pytest
2. Build Docker image: docker build .
3. Push to registry
4. Deploy to Kubernetes cluster
5. Monitor logs for errors"""
}

for path, content in sample_docs.items():
    add_document(path, content)


if __name__ == "__main__":
    print(f"Knowledge base: {len(documents)} documents")
    mcp.run()