Context Bridge 🌉

Unified Python package for RAG-powered documentation management - Crawl, store, chunk, and retrieve technical documentation with vector + BM25 hybrid search.

---

📋 Table of Contents

• Overview
• Features
• Architecture
• Installation
• Quick Start
• Configuration
• Usage
• Database Schema
• Development
• Technical Documentation
• License

---

🎯 Overview

Context Bridge is a standalone Python package designed to help AI agents, LLMs, and developers manage technical documentation with RAG (Retrieval-Augmented Generation) capabilities. It bridges the gap between scattered online documentation and AI-ready, searchable knowledge bases.

What It Does

1. Crawls technical documentation from URLs using Crawl4AI
2. Organizes crawled pages into logical groups with size management
3. Chunks Markdown content intelligently while preserving structure
4. Embeds chunks using vector embeddings (Ollama/Gemini)
5. Stores everything in PostgreSQL with vector and vchord_bm25
6. Searches with hybrid vector + BM25 search for best results
7. Serves via MCP (Model Context Protocol) for AI agent integration
8. Manages through a Streamlit UI for human oversight

---

✨ Features

Core Capabilities

• 🕷️ Smart Crawling: Automatically detect and crawl documentation sites, sitemaps, and text files
• 📦 Intelligent Chunking: Smart Markdown chunking that respects code blocks, paragraphs, and sentences
• 🔍 Hybrid Search: Dual vector + BM25 search for superior retrieval accuracy
• 📚 Version Management: Track multiple versions of the same documentation
• 🎯 Document Organization: Manual page grouping with size constraints before chunking
• ⚡ High Performance: PSQLPy for fast async PostgreSQL operations
• 🤖 AI-Ready: MCP server for seamless AI agent integration
• 🎨 User-Friendly: Streamlit UI for documentation management

Technical Features

• Vector Search: Powered by vector extension
• BM25 Full-Text Search: Using vchord_bm25 extension
• Async/Await: Fully asynchronous operations for scalability
• Configurable Embeddings: Support for Ollama (local) and Google Gemini (cloud)
• Type-Safe: Pydantic models for configuration and data validation
• Modular Design: Clean separation of concerns (repositories, services, managers)

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────────┐
│                    Context Bridge Architecture               │
└─────────────────────────────────────────────────────────────┘

┌──────────────┐    ┌──────────────┐    ┌──────────────┐
│  Streamlit   │    │ MCP Server   │    │  Python API  │
│      UI      │    │  (AI Agent)  │    │   (Direct)   │
└──────┬───────┘    └──────┬───────┘    └──────┬───────┘
       │                   │                   │
       └───────────────────┴───────────────────┘
                           │
                           ▼
       ┌───────────────────────────────────────┐
       │         Service Layer                 │
       │  - CrawlingService                    │
       │  - ChunkingService                    │
       │  - EmbeddingService                   │
       │  - SearchService                      │
       └───────────────┬───────────────────────┘
                       │
                       ▼
       ┌───────────────────────────────────────┐
       │       Repository Layer                │
       │  - DocumentRepository                 │
       │  - PageRepository                     │
       │  - GroupRepository                    │
       │  - ChunkRepository                    │
       └───────────────┬───────────────────────┘
                       │
                       ▼
       ┌───────────────────────────────────────┐
       │      PostgreSQL Manager               │
       │  - Connection Pooling                 │
       │  - Transaction Management             │
       └───────────────┬───────────────────────┘
                       │
                       ▼
       ┌───────────────────────────────────────┐
       │     PostgreSQL Database               │
       │  Extensions:                          │
       │  - vector (vector search)           │
       │  - vchord_bm25 (BM25 search)        │
       │  - pg_tokenizer (text tokenization) │
       └───────────────────────────────────────┘

External Dependencies:
┌──────────────┐    ┌──────────────┐
│   Crawl4AI   │    │    Ollama    │
│  (Crawling)  │    │  or Gemini   │
│              │    │ (Embeddings) │
└──────────────┘    └──────────────┘

Workflow

1. Crawl Documentation
   ↓
2. Store Raw Pages
   ↓
3. Manual Organization (Group Pages)
   ↓
4. Smart Chunking
   ↓
5. Generate Embeddings
   ↓
6. Store with Vector + BM25 Indexes
   ↓
7. Hybrid Search (Vector + BM25)

---

📦 Installation

Prerequisites

• Python 3.11+
• PostgreSQL 14+ with extensions:
  • vector
  • vchord
  • pg_tokenizer
  • vchord_bm25
• Ollama (for local embeddings) or Google API Key (for Gemini)

Install from PyPI

pip install context-bridge

Install with Optional Dependencies

# With Gemini support
pip install context-bridge[gemini]

# With MCP server
pip install context-bridge[mcp]

# With Streamlit UI
pip install context-bridge[ui]

# All features
pip install context-bridge[all]

Running the Applications

MCP Server:

# Using the installed script
context-bridge-mcp

# Or run directly
python -m context_bridge_mcp

Streamlit UI:

# Using streamlit directly
streamlit run streamlit_app/app.py

# Or with uv
uv run streamlit run streamlit_app/app.py

Install from Source

git clone https://github.com/yourusername/context-bridge.git
cd context-bridge
pip install -e .

---

🚀 Quick Start

1. Initialize Database

python -m context_bridge.database.init_databases

This will:

• Create required PostgreSQL extensions
• Create all necessary tables
• Set up vector and BM25 indexes

2. Basic Usage (Three Ways)

Option A: Direct Python (Recommended for PyPI users)

import asyncio
from context_bridge import ContextBridge, Config

async def main():
    # Create config with your settings
    config = Config(
        postgres_host="localhost",
        postgres_password="your_secure_password",
        embedding_model="nomic-embed-text:latest"
    )
    
    # Use with context manager
    async with ContextBridge(config=config) as bridge:
        # Crawl documentation
        result = await bridge.crawl_documentation(
            name="Python Docs",
            version="3.11",
            source_url="https://docs.python.org/3/library/"
        )
        
        # Search documentation
        search_results = await bridge.search(
            query="async await tutorial",
            document_id=result.document_id
        )
        
        for hit in search_results[:3]:
            print(f"Score: {hit.score}, Content: {hit.content[:100]}...")

if __name__ == "__main__":
    asyncio.run(main())

Option B: Environment Variables (Recommended for Docker/K8s)

# Set environment variables
export POSTGRES_HOST=postgres
export POSTGRES_PASSWORD=secure_password
export OLLAMA_BASE_URL=http://ollama:11434
export EMBEDDING_MODEL=nomic-embed-text:latest

# Or in docker-compose.yml
environment:
  - POSTGRES_HOST=postgres
  - POSTGRES_PASSWORD=secure_password
  - EMBEDDING_MODEL=nomic-embed-text:latest

import asyncio
from context_bridge import ContextBridge

async def main():
    # Config automatically loaded from environment variables
    async with ContextBridge() as bridge:
        result = await bridge.crawl_documentation(
            name="Python Docs",
            version="3.11",
            source_url="https://docs.python.org/3/library/"
        )

Option C: .env File (Convenient for local development)

Create .env file (git-ignored):

# PostgreSQL Configuration
POSTGRES_HOST=localhost
POSTGRES_PORT=5432
POSTGRES_USER=postgres
POSTGRES_PASSWORD=your_secure_password
POSTGRES_DB=context_bridge

# Ollama Configuration
OLLAMA_BASE_URL=http://localhost:11434
EMBEDDING_MODEL=nomic-embed-text:latest
VECTOR_DIMENSION=768

# Search Configuration
SIMILARITY_THRESHOLD=0.7
BM25_WEIGHT=0.3
VECTOR_WEIGHT=0.7

# Chunking Configuration
CHUNK_SIZE=2000
MIN_COMBINED_CONTENT_SIZE=100
MAX_COMBINED_CONTENT_SIZE=3500000

# Crawling Configuration
CRAWL_MAX_DEPTH=3
CRAWL_MAX_CONCURRENT=5

Then in your code:

import asyncio
from context_bridge import ContextBridge

async def main():
    # Config automatically loaded from .env file (if python-dotenv is available)
    async with ContextBridge() as bridge:
        result = await bridge.crawl_documentation(...)

To use .env files in development, install with dev dependencies:

pip install context-bridge[dev]

---

⚙️ Configuration

The package uses Pydantic for type-safe, type-hinted configuration. Context Bridge supports three configuration methods:

Configuration Methods (Priority Order)

1. Direct Python instantiation (recommended for packaged installs)
2. Environment variables (recommended for containers/CI)
3. .env file (convenient for local development only)

Core Settings

Setting	Default	Description	Python API
POSTGRES_HOST	localhost	PostgreSQL host	postgres_host
POSTGRES_PORT	5432	PostgreSQL port	postgres_port
POSTGRES_USER	postgres	PostgreSQL user	postgres_user
POSTGRES_PASSWORD	`` (empty)	PostgreSQL password (min 8 chars for prod)	postgres_password
POSTGRES_DB	context_bridge	Database name	postgres_db
DB_POOL_MAX	10	Connection pool size	postgres_max_pool_size

Embedding Settings

Setting	Default	Description	Python API
OLLAMA_BASE_URL	http://localhost:11434	Ollama API URL	ollama_base_url
EMBEDDING_MODEL	nomic-embed-text:latest	Ollama model name	embedding_model
VECTOR_DIMENSION	768	Embedding vector dimension	vector_dimension

Search Settings

Setting	Default	Description	Python API
SIMILARITY_THRESHOLD	0.7	Minimum similarity score	similarity_threshold
BM25_WEIGHT	0.3	BM25 weight in hybrid search	bm25_weight
VECTOR_WEIGHT	0.7	Vector weight in hybrid search	vector_weight

Chunking Settings

Setting	Default	Description	Python API
CHUNK_SIZE	2000	Default chunk size (bytes)	chunk_size
MIN_COMBINED_CONTENT_SIZE	100	Minimum combined page size (bytes)	min_combined_content_size
MAX_COMBINED_CONTENT_SIZE	3500000	Maximum combined page size (bytes)	max_combined_content_size

Crawling Settings

Setting	Default	Description	Python API
CRAWL_MAX_DEPTH	3	Maximum crawl depth	crawl_max_depth
CRAWL_MAX_CONCURRENT	10	Maximum concurrent crawl operations	crawl_max_concurrent

---

📚 Usage

Crawling Documentation

from context_bridge.service.crawling_service import CrawlingService, CrawlConfig
from crawl4ai import AsyncWebCrawler

# Configure crawler
config = CrawlConfig(
    max_depth=3,          # How deep to follow links
    max_concurrent=10,    # Concurrent requests
    memory_threshold=70.0 # Memory usage threshold
)

service = CrawlingService(config)

# Crawl a documentation site
async with AsyncWebCrawler(verbose=True) as crawler:
    result = await service.crawl_webpage(
        crawler,
        "https://docs.example.com"
    )

# Access results
for crawl_result in result.results:
    print(f"URL: {crawl_result.url}")
    print(f"Content length: {len(crawl_result.markdown)}")

Storing Documents

from context_bridge.repositories.document_repository import DocumentRepository

async with db_manager.connection() as conn:
    doc_repo = DocumentRepository(conn)
    
    # Create a new document
    doc_id = await doc_repo.create(
        name="Python Documentation",
        version="3.11",
        source_url="https://docs.python.org/3/",
        description="Official Python 3.11 documentation"
    )
    
    # Store crawled pages
    for page in crawled_pages:
        await page_repo.create(
            document_id=doc_id,
            url=page.url,
            content=page.markdown,
            content_hash=hash(page.markdown)
        )

Organizing Pages into Groups

from context_bridge.repositories.group_repository import GroupRepository

# User manually selects pages to group
page_ids = [1, 2, 3, 4, 5]

# Create a group
async with db_manager.connection() as conn:
    group_repo = GroupRepository(conn)
    
    group_id = await group_repo.create_group(
        document_id=doc_id,
        page_ids=page_ids,
        min_size=1000,   # Minimum total content size
        max_size=50000   # Maximum total content size
    )

Chunking and Embedding

from context_bridge.service.chunking_service import ChunkingService
from context_bridge.service.embedding import EmbeddingService

chunking_service = ChunkingService()
embedding_service = EmbeddingService(config)

# Get groups ready for chunking
eligible_groups = await group_repo.get_eligible_groups(doc_id)

for group in eligible_groups:
    # Get combined content
    content = await group_repo.get_group_content(group.id)
    
    # Smart chunking
    chunks = chunking_service.smart_chunk_markdown(
        content,
        chunk_size=2000
    )
    
    # Generate embeddings and store
    for i, chunk_text in enumerate(chunks):
        embedding = await embedding_service.get_embedding(chunk_text)
        
        await chunk_repo.create(
            document_id=doc_id,
            group_id=group.id,
            chunk_index=i,
            content=chunk_text,
            embedding=embedding
        )

Searching Documents

Find Documents by Query

from context_bridge.repositories.document_repository import DocumentRepository

# Find relevant documents
documents = await doc_repo.find_by_query(
    query="python asyncio tutorial",
    limit=5
)

for doc in documents:
    print(f"{doc.name} (v{doc.version})")

Search Document Content (Hybrid Search)

from context_bridge.repositories.chunk_repository import ChunkRepository

# Search within a specific document
chunks = await chunk_repo.hybrid_search(
    document_id=doc_id,
    version="3.11",
    query="async await examples",
    query_embedding=await embedding_service.get_embedding("async await examples"),
    limit=10,
    vector_weight=0.7,
    bm25_weight=0.3
)

for chunk in chunks:
    print(f"Score: {chunk.score}")
    print(f"Content: {chunk.content[:200]}...")

Using the Streamlit UI

The Context Bridge includes a full-featured web interface for managing documentation:

# Install with UI support
pip install context-bridge[ui]

# Run the Streamlit application
uv run streamlit run streamlit_app/app.py

# Or use the installed script
context-bridge-ui

Features:

• Document Management: View, search, and delete documents
• Page Organization: Select and group crawled pages for processing
• Chunk Processing: Convert page groups into searchable chunks
• Hybrid Search: Search across all documentation with advanced filtering

Using the MCP Server

The Model Context Protocol server allows AI agents to interact with Context Bridge:

# Install with MCP support
pip install context-bridge[mcp]

# Run the MCP server
uv run python -m context_bridge_mcp

# Or use the installed script
context-bridge-mcp

Available Tools:

• find_documents: Search for documents by query
• search_content: Perform hybrid vector + BM25 search within specific documents

Integration with AI Clients: The MCP server can be integrated with AI assistants like Claude Desktop for seamless documentation access.

For detailed usage instructions, see the MCP Server Usage Guide.

---

🗄️ Database Schema

Core Tables

-- Documents (versioned documentation)
CREATE TABLE documents (
    id SERIAL PRIMARY KEY,
    name TEXT NOT NULL,
    version TEXT NOT NULL,
    source_url TEXT,
    description TEXT,
    metadata JSONB DEFAULT '{}'::jsonb,
    created_at TIMESTAMPTZ DEFAULT NOW(),
    updated_at TIMESTAMPTZ DEFAULT NOW(),
    UNIQUE(name, version)
);

-- Pages (raw crawled content)
CREATE TABLE pages (
    id SERIAL PRIMARY KEY,
    document_id INTEGER NOT NULL REFERENCES documents(id) ON DELETE CASCADE,
    url TEXT NOT NULL UNIQUE,
    content TEXT NOT NULL,
    content_hash TEXT NOT NULL,
    content_length INTEGER GENERATED ALWAYS AS (length(content)) STORED,
    crawled_at TIMESTAMPTZ DEFAULT NOW(),
    status TEXT DEFAULT 'pending' CHECK (status IN ('pending', 'processing', 'chunked', 'deleted')),
    group_id UUID, -- For future grouping feature
    metadata JSONB DEFAULT '{}'::jsonb
);

-- Chunks (embedded content)
CREATE TABLE chunks (
    id SERIAL PRIMARY KEY,
    document_id INTEGER NOT NULL REFERENCES documents(id) ON DELETE CASCADE,
    chunk_index INTEGER NOT NULL,
    content TEXT NOT NULL,
    group_id UUID, -- For future grouping feature
    embedding VECTOR(768), -- Dimension must match config
    bm25_vector bm25vector, -- Auto-generated by trigger
    created_at TIMESTAMPTZ DEFAULT NOW(),
    UNIQUE(document_id, group_id, chunk_index)
);

-- Indexes
CREATE INDEX idx_pages_document ON pages(document_id);
CREATE INDEX idx_pages_status ON pages(status);
CREATE INDEX idx_pages_hash ON pages(content_hash);
CREATE INDEX idx_pages_group ON pages(group_id);
CREATE INDEX idx_chunks_document ON chunks(document_id);
CREATE INDEX idx_chunks_group ON chunks(group_id);
CREATE INDEX idx_chunks_vector ON chunks USING ivfflat(embedding vector_cosine_ops) WITH (lists = 100);
CREATE INDEX idx_chunks_bm25 ON chunks USING bm25(bm25_vector bm25_ops);

---

🛠️ Development

Project Structure

context_bridge/               # Core package
├── __init__.py
├── config.py                 # Configuration management
├── core.py                   # Main ContextBridge API
├── database/
│   ├── init_databases.py     # Database initialization
│   └── postgres_manager.py   # Connection pool manager
├── schema/
│   └── extensions.sql        # PostgreSQL extensions & schema
├── repositories/             # Data access layer
│   ├── document_repository.py
│   ├── page_repository.py
│   ├── group_repository.py
│   └── chunk_repository.py
├── service/                  # Business logic layer
│   ├── crawling_service.py
│   ├── chunking_service.py
│   ├── embedding.py
│   ├── search_service.py
│   └── url_service.py

context_bridge_mcp/          # MCP Server (Model Context Protocol)
├── __init__.py
├── server.py                 # MCP server implementation
├── schemas.py                # Tool input/output schemas
└── __main__.py               # CLI entry point

streamlit_app/               # Streamlit Web UI
├── __init__.py
├── app.py                    # Main application
├── pages/                    # Multi-page navigation
│   ├── documents.py          # Document management
│   ├── crawled_pages.py      # Page management
│   └── search.py             # Search interface
├── components/               # Reusable UI components
├── utils/                    # UI utilities and helpers
└── README.md                 # UI-specific documentation

docs/                        # Documentation
├── guide/
│   └── MCP_SERVER_USAGE.md   # MCP server usage guide
├── plan/                    # Development plans
│   └── ui_and_mcp_implementation_plan.md
├── technical/               # Technical guides
│   ├── crawl4ai_complete_guide.md
│   ├── embedding_service.md
│   ├── psqlpy-complete-guide.md
│   ├── python_mcp_server_guide.md
│   ├── python-testing-guide.md
│   └── smart_chunk_markdown_algorithm.md
└── memory_templates.yaml    # Memory usage templates

tests/                       # Test suite
├── conftest.py
├── integration/
├── unit/
└── e2e/                     # End-to-end tests
    ├── conftest.py
    └── test_streamlit_ui.py

### Running Tests

```bash
# Install dev dependencies
pip install -e ".[dev]"

# Run all tests
pytest

# Run with coverage
pytest --cov=context_bridge --cov-report=html

# Run specific test file
pytest tests/test_chunking_service.py -v

Code Quality

# Format code
black context_bridge tests

# Type checking
mypy context_bridge

# Linting
ruff check context_bridge

---

📖 Technical Documentation

Comprehensive technical guides are available in docs/:

Testing & Quality Assurance

• UI Testing Report - Comprehensive Playwright testing results and bug fixes
• MCP Server Usage Guide - How to use the MCP server with AI clients

Technical Guides (docs/technical/)

• Crawl4AI Guide - Complete crawling documentation
• Embedding Service - Ollama and Gemini embedding setup
• PSQLPy Guide - PostgreSQL driver usage
• MCP Server Guide - MCP server implementation
• Testing Guide - Testing best practices
• Smart Chunking Algorithm - Chunking implementation

Implementation Plans (docs/plan/)

• UI & MCP Implementation Plan - Development roadmap and progress

---

🤝 Contributing

Contributions are welcome! Please read our contributing guidelines and submit pull requests.

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

---

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

---

🙏 Acknowledgments

• Crawl4AI - High-performance web crawler
• PSQLPy - Async PostgreSQL driver
• pgvector - Vector similarity search
• MCP - Model Context Protocol

---

📧 Support

For questions, issues, or feature requests:

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Email: your.email@example.com

---

Built with ❤️ for AI agents and developers