Simple Files Vectorstore

You can search your files semantically by watching specified directories, generating vector embeddings from their contents, and querying them with natural language queries. This MCP server keeps your documents up to date as files change, enabling fast, relevant search across your corpus.

How to use

To perform semantic searches, connect your MCP client to the simple-files-vectorstore server and issue a search query. The server returns relevant results with matched content, source file paths, and similarity scores. You can also request statistics to understand how many documents are indexed and which directories are being watched.

How to install

Prerequisites: install Node.js and npm on your machine.

Add the MCP server configuration to your MCP settings. Use the following snippet to start watching your directories and run the vector store locally with npx.

{
  "mcpServers": {
    "files_vectorstore": {
      "command": "npx",
      "args": [
        "-y",
        "@lishenxydlgzs/simple-files-vectorstore"
      ],
      "env": {
        "WATCH_DIRECTORIES": "/path/to/your/directories"
      },
      "disabled": false,
      "autoApprove": []
    }
  }
}

Additional configuration and usage notes

Configuration is driven by environment variables. You must specify the directories to watch using either WATCH_DIRECTORIES or WATCH_CONFIG_FILE.

Watch directories: Provide a comma-separated list of directories to monitor. If you prefer a JSON config file, point WATCH_CONFIG_FILE to a file that contains a watchList array.

Optional parameters allow you to tune processing: adjust CHUNK_SIZE for how many characters are processed per chunk, CHUNK_OVERLAP to control overlap between chunks, and IGNORE_FILE to exclude certain files or directories based on patterns.

{
    "mcpServers": {
      "files-vectorstore": {
        "command": "npx",
        "args": [
          "-y",
          "@lishenxydlgzs/simple-files-vectorstore"
        ],
        "env": {
          "WATCH_DIRECTORIES": "/path/to/dir1,/path/to/dir2",
          "CHUNK_SIZE": "2000",
          "CHUNK_OVERLAP": "500",
          "IGNORE_FILE": "/path/to/.gitignore"
        },
        "disabled": false,
        "autoApprove": []
      }
    }
  }

You can also use a watch config file by setting WATCH_CONFIG_FILE to the path of a JSON file that contains a watchList. The watchList is an array of paths you want to monitor.

{
  "watchList": [
    "/path/to/dir1",
    "/path/to/dir2",
    "/path/to/specific/file.txt"
  ]
}

Available tools

search

Perform semantic search across indexed files by querying the embeddings and retrieving matching content with scores and sources.

get_stats

Return statistics about indexed documents, watched directories, and currently processing files.