User / MCP Client
       |
       v
MCP Server (FastMCP)
       |
       +----------------------+
       |                      |
       v                      v
 search_notes()        get_document()
       |
       v
Gemini Embedding API
       |
       v
Qdrant Vector Database
       |
       v
Ranked Chunks
       |
       v
Source + Page + Score + Text
## Features

* PDF document ingestion
* Page-by-page text extraction
* Recursive text chunking
* Gemini `gemini-embedding-001` embeddings
* Qdrant vector storage
* Semantic similarity search
* Source and page citations
* Confidence threshold for low-relevance queries
* Full-document retrieval
* Indexed-source listing
* MCP Inspector support

## MCP Tools

### `search_notes`

Searches the knowledge base using semantic similarity.

Arguments:

* `query`: search question or topic
* `top_k`: maximum number of results

Returns:

* similarity score
* source filename
* page number
* relevant text chunk

### `get_document`

Returns the complete text of an indexed PDF document.

Argument:

* `doc_id`: document filename

Example:

```text
Complex_Variables_Project_Report.pdf

1. Complex_Variables_Project_Report.pdf

Personal-Knowledge-MCP/
├── documents/
│   └── Complex_Variables_Project_Report.pdf
├── services/
│   ├── chunking.py
│   ├── embedding.py
│   ├── pdf_reader.py
│   └── qdrant_service.py
├── .env
├── .gitignore
├── evaluation.py
├── ingest.py
├── requirements.txt
└── server.py

python -m venv .venv
.venv\Scripts\Activate.ps1

pip install -r requirements.txt

GEMINI_API_KEY=your_api_key_here

http://localhost:6333

docker run -d --name qdrant -p 6333:6333 -p 6334:6334 qdrant/qdrant

documents/

python ingest.py

PDF
 ↓
Page extraction
 ↓
Chunking
 ↓
Gemini embeddings
 ↓
Qdrant storage

text
page
source

mcp dev server.py

search_notes
get_document
list_source_documents

Tests: 5
Successful hits: 5
Hit@3: 100%

0.60

0.79
0.76
0.75

0.52

No confident match found.

Complex_Variables_Project_Report.pdf