Personal Knowledge-Base MCP Server

A recruiter-ready Model Context Protocol (MCP) server that provides semantic search over a real, personally-owned document corpus.

The project combines FastMCP, Qdrant vector search, sentence-transformer embeddings, document ingestion, retrieval evaluation, FastAPI, SQLite authentication, and a multi-user web interface.

1. Overview

Traditional keyword search can miss relevant information when the wording of a query differs from the wording used in a document.

This project solves that problem by converting documents into vector embeddings and storing them in Qdrant. Users can then search their knowledge base by meaning rather than exact keywords.

The retrieval functionality is exposed through the Model Context Protocol (MCP), allowing an MCP-compatible client to call the knowledge-base tools directly.

The project is designed as a working, demo-able retrieval server rather than a chatbot application.

Related MCP server: mycelium

2. Problem Statement

Static keyword search depends heavily on matching exact words.

For example, a document may contain:

"Agents receive percepts from their environment and perform rational actions."

while a user searches:

"What is an intelligent agent?"

A semantic retrieval system can recognize that these concepts are related even when the wording is different.

This project provides reusable semantic retrieval tools backed by a vector database instead of building a one-off chatbot.

3. Goals

The project was designed to:

• Build a working MCP server exposing reusable retrieval tools.

• Index a real student-owned document corpus.

• Ingest PDF, Markdown, and TXT documents.

• Chunk documents before embedding.

• Generate semantic embeddings.

• Store embeddings in Qdrant.

• Retrieve semantically relevant document chunks.

• Return ranked results with source citations.

• Retrieve complete indexed document context.

• List indexed sources.

• Support per-user document isolation.

• Return no_confident_match for low-confidence queries.

• Measure retrieval quality using hand-labeled queries.

• Provide a lightweight multi-user web demonstration.

• Provide a live, connectable MCP server suitable for an MCP-compatible client.

4. Target Corpus

The project uses a real, student-owned corpus rather than a generic tutorial or Kaggle dataset.

The current demonstration corpus contains:

• Samavia_Semester_Notes.md

The notes contain Artificial Intelligence course material including:

• Intelligent Agents

• Environment Types

• Agent Architectures

• Search Algorithms

• Breadth-First Search

• A* Search

• Greedy Best-First Search

• Minimax

• Alpha-Beta Pruning

• Propositional Logic

• First-Order Logic

• Reinforcement Learning

• Bayes' Theorem

• Perceptrons

• Backpropagation

Additional personal documents can be uploaded through the web interface in PDF, Markdown, or TXT format.

5. Architecture

                     ┌──────────────────────┐
                     │     MCP Client       │
                     │ Claude / MCP Client  │
                     └──────────┬───────────┘
                                │
                         MCP / JSON-RPC
                                │
                                ▼
                     ┌──────────────────────┐
                     │    FastMCP Server    │
                     │                      │
                     │  search_notes()      │
                     │  get_document()      │
                     │  list_sources()      │
                     └──────────┬───────────┘
                                │
                                ▼
                     ┌──────────────────────┐
                     │   Retrieval Core     │
                     │                      │
                     │ Chunking             │
                     │ Embeddings           │
                     │ Similarity Search    │
                     │ Threshold Filtering  │
                     └──────────┬───────────┘
                                │
                                ▼
                     ┌──────────────────────┐
                     │    Qdrant Cloud      │
                     │   Vector Database    │
                     └──────────────────────┘


        ┌─────────────────────────────────────────┐
        │             Web Application             │
        │                                         │
        │  Sign Up / Login                        │
        │  Document Upload                        │
        │  Semantic Search                        │
        │  Source Listing                         │
        │  Search Results                         │
        └──────────────────┬──────────────────────┘
                           │
                           ▼
                    ┌───────────────┐
                    │   FastAPI     │
                    │   Backend     │
                    └───────┬───────┘
                            │
                            ▼
                     Retrieval Core