MCP Enhanced Data Retrieval System

An MCP (Model Context Protocol) server that standardizes AI context sharing by integrating organizational knowledge sources (GitHub, internal docs, APIs) to enable domain-aware AI assistance for enterprise development workflows.

Project Overview

This system implements the Model Context Protocol to provide:

• Standardized AI context sharing across organizational knowledge sources

• GitHub repository integration with OAuth 2.1 authentication

• Vector-based semantic search using embeddings

• Optimized 1500-token context chunking for sub-500ms TTFT

• Parallel retrieval strategy with 2-second timeout

• Streamable HTTP transport using FastAPI

Architecture

Features

• MCP Protocol Compliance: JSON-RPC 2.0 over Streamable HTTP

• GitHub Integration: Repository data retrieval and contextualization

• Vector Embeddings: Semantic search using ChromaDB and Sentence Transformers

• Context Optimization: 1500-token chunking with parallel retrieval

• OAuth 2.1 Security: Secure authentication for GitHub access

• Performance: Sub-500ms response times with 2-second retrieval timeout

Project Structure

Setup

1. Clone and navigate to the project:

   cd "MCP Enhanced Data Retrieval"

2. Create virtual environment:

   python3 -m venv venv source venv/bin/activate # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Configure environment variables:

   cp .env.example .env # Edit .env with your credentials

5. Run the server:

   uvicorn src.server.main:app --reload

Milestone 1 Goals

• ✅ MCP protocol analysis and communication flow evaluation

• ✅ High-level architecture design for enterprise knowledge integration

• 🔄 Functional MCP server with GitHub integration

• 🔄 OAuth 2.1 authentication implementation

• 🔄 1500-token context chunking mechanism

• 🔄 Vector-based semantic search

Success Criteria

• Functional MCP server that can retrieve and contextualize GitHub repository information

• OAuth 2.1 authentication for secure GitHub access

• 1500-token context chunking maintaining sub-500ms TTFT

• Parallel retrieval with 2-second timeout

• Vector-based semantic search for relevant content

Technologies

• MCP SDK: Anthropic MCP Python SDK

• Web Framework: FastAPI with Streamable HTTP transport

• GitHub API: PyGithub

• Authentication: OAuth 2.1 (authlib)

• Vector Database: ChromaDB

• Embeddings: Sentence Transformers (all-MiniLM-L6-v2)

• Token Processing: tiktoken

Author

Kalpalathika Ramanujam Advisor: Dr. Thomas Kinsman Rochester Institute of Technology

License

Academic Project - RIT Capstone