mcp-docs-reader
About
A lightweight MCP server that loads PDF files, extracts and chunks their text, builds a semantic vector index, and returns relevant passages to Claude or other AI agents for document-based question answering.
Details
- Author
- AIMIZING
- Downloads
- 214
- Categories
- Knowledge Base
Jump to
- Loads and processes PDF documents from a local docs/ folder
- Extracts text and splits it into semantic chunks
- Generates vector embeddings using SentenceTransformer
- Builds a FAISS-based vector index for semantic search
- Retrieves top-k relevant chunks based on user query
- Constructs a prompt with passages and returns it to Claude
Setting up with Highlight
This MCP is not yet compatible with Highlight’s one-click setup. However, you can still use it with Highlight by following these steps:
- Download and install Highlight from highlightai.com/download
- Navigate to the plugins tab and select "Add Custom Plugin"
-
Configure the plugin with the settings below
Plugin Name
mcp-docs-readerCommand (node, npx, python, etc.)Please refer to the README for specific instructions on how to obtain API keys or other required environment variables.
- Enable "Start Automatically" if you want the plugin to start when Highlight launches
From the repository
Download or clone the project, set up the uv virtual environment (manually or via setup.bat), and install dependencies. Then edit your Claude Desktop claude_desktop_config.json to include the content from docReader_config.json, replacing the placeholder path with the actual project path. Launch Claude Desktop; it will automatically connect to the MCP tool.
Claude Desktop / Cursor
Paste into your MCP client config file to install this server.
{
"mcpServers": {
"mcp-docs-reader": {
"mcp_docs_reader": {
"command": "uv",
"args": [
"venv"
]
}
}
}
}McpServers
{
"mcp_docs_reader": {
"command": "uv",
"args": [
"venv"
]
}
}mcp-docs-reader
A lightweight MCP (Model Context Protocol) server that loads PDF files from a local folder, extracts and chunks their content, builds a semantic search index, and sends relevant passages to Claude Desktop for document-based question answering.This project is intended to be used with Claude's MCP desktop feature.
---
Features
- Loads and processes PDF documents from a local docs/ folder - Extracts text and splits it into semantic chunks - Generates vector embeddings using SentenceTransformer - Builds a FAISS-based vector index for semantic search - Retrieves top-k relevant chunks based on user query - Constructs a prompt (relevant passages + question) and returns it to Claude - Minimal setup using uv---
How to Use
1. Install Claude Desktop (if not installed)
- Download Claude Desktop and install it.2. Download mcp-docs-reader
Option 1: Download as ZIP
- Click "Code" > "Download ZIP"
- Extract the downloaded ZIP file
Option 2: Clone with Git
git clone https://github.com/AIMIZING/mcp_docs_reader.git
cd mcp_docs_reader
3. Set Up UV Environment
Option 1: Manual Setup - Install uv (if not installed):powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
uv venv
.venv\Scripts\activate
pip install -r requirements.txt
Option 2: Automatic Setup (Windows)
- run the included "setup.bat" file
4. Configure Claude Desktop
Open the included "weather_config.json" file, and copy its content. Then, open your existing Claude Desktop configuration file and append the copied content.To locate the Claude Desktop config file:
- Open Claude Desktop
- Go to Menu → File → Settings → Developer Mode → Edit Configuration
- This will open your current "claude_desktop_config.json" file
- Paste the additional content at the appropriate position (e.g. within the mcp list or relevant section)
⚠️ Before copying, make sure to replace the path
"C:\\PATH\\TO\\mcp_docs_reader" in docReader_config.json
with your actual local project path (e.g. C:\\Users\\YourName\\Documents\\mcp_docs_reader).
⚠️ Do not overwrite the entire file — make sure to append or merge the content to avoid breaking existing configurations.
5. Run Claude Desktop
Launch Claude Desktop. Once it's running, it will automatically detect and connect to the configured MCP tool. You can now ask Claude questions based on your local PDF documents, like:> "Summarize the key points from the registered file contents."
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.
