home / mcp / mcp vectorstore server

MCP VectorStore Server

This MCP document Server with an integrated Faiss vectorstore works with popular MCP clients like GitHub Copilot and VSCode’s Continue extension — all running smoothly on Ubuntu WSL for Windows users.

Installation
Add the following to your MCP client configuration file.

Configuration

View docs
{
  "mcpServers": {
    "bierschneideremanuel-mcpdocserver": {
      "command": "python",
      "args": [
        "/home/em/McpDocServer/mcp_vectorstore_server.py"
      ],
      "env": {
        "PYTHONPATH": "/home/em/McpDocServer/McpDocServer",
        "USER_AGENT": "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/119.0.0.0 Safari/537.36",
        "VECTORSTORE_DIR": "/path/to/your/documents",
        "LLMSHERPA_API_URL": "http://localhost:5001/api/parseDocument?renderFormat=all",
        "CUDA_VISIBLE_DEVICES": "0"
      }
    }
  }
}

You run an MCP VectorStore Server to manage document vector stores, process PDFs, and perform semantic searches. It exposes a standardized MCP interface so you can search, create, and manage vector stores, read local files, perform web searches, and even calculate mathematical expressions from a unified client workflow.

How to use

Start by launching the local MCP VectorStore Server. Use a Python environment and run the server script so your MCP clients can connect. The server provides tools to search, create, inspect, and clear vector stores, as well as read local files and perform web searches. You can integrate it with clients that support MCP to issue tool calls like vectorstore_search, vectorstore_create, and vectorstore_info. When you query, you receive relevant document snippets from indexed stores and can perform additional operations such as reading files or calculating expressions.

How to install

Prerequisites: ensure you have Python 3.8 or higher installed on your system and a supported operating system (Linux, macOS, or Windows). You should have at least 4 GB RAM (8 GB+ recommended for large collections) and several gigabytes of free storage for models and vector stores.

Step 1: Create a working directory for the server and its data, then navigate into it.

Step 2: Create and activate a virtual environment.

Step 3: Install dependencies from the requirements file.

Step 4: (Optional but recommended) Install LLMSherpa for PDF processing and start its server if you will index and extract content from PDFs.

Step 5: Prepare your document directory with subfolders for PDFs and text files that you plan to index.

Configuration and startup

To run the server locally, use the following startup command as shown in the example usage. This starts the MCP VectorStore Server in the local environment.

python /home/em/McpDocServer/mcp_vectorstore_server.py

Client integration examples

Configure your MCP client to connect to the local server using a stdio connection. The following example shows how to wire the server into a client configuration. This enables you to issue tool calls to create, search, and manage vector stores directly from your MCP-enabled client.

{
  "mcpServers": {
    "vectorstore": {
      "command": "python",
      "args": ["/home/em/McpDocServer/mcp_vectorstore_server.py"],
      "env": {
        "PYTHONPATH": "/home/em/McpDocServer/McpDocServer"
      }
    }
  }
}

Available tools

The server exposes a set of tools to interact with the vector store, read files, perform web searches, and compute calculations.

Available tools

vectorstore_search

Search the vector store for relevant documents using a query and a result count.

vectorstore_create

Create a new vector store from documents found in a specified directory.

vectorstore_info

Retrieve information about the current vector store, such as document count and model details.

vectorstore_clear

Remove all documents from the current vector store.

read_file

Read and return the contents of a file on the local system.

google_search

Perform a Google search for a given query and return results.

wikipedia_search

Search Wikipedia for information related to a query.

duckduckgo_search

Perform a DuckDuckGo search for information.

calculate

Evaluate a mathematical expression and return the result.