Simple Files Vectorstore

Simple Files Vectorstore is an MCP server that provides semantic search across local documents by creating vector embeddings from watched directories. It connects an AI assistant to local file systems, making it useful for software engineers, researchers, and technical writers who want to perform retrieval-augmented generation directly on their desktop environments. Once configured, the server actively monitors specified directories for additions, updates, or deletions, automatically keeping an indexed vector store synchronized in the background. Users can search their local files using natural language queries rather than exact string matches, retrieving contextually relevant text passages alongside source paths and relevance scores. The server also offers administrative tools to query current indexing status, such as active processing files and total document counts. By exposing standard Model Context Protocol tools, Simple Files Vectorstore enables clients to retrieve local text chunks dynamically during conversations, improving accuracy when analyzing project documentation, codebases, and personal knowledge repositories.

Category: AI Memory & Context

Tags: embeddings, local-files, rag, semantic search, vector-search

Visit Simple Files Vectorstore

How to install and configure Simple Files Vectorstore

To install and run Simple Files Vectorstore with an MCP client like Claude Desktop or VSCode Cline, add the server entry to your configuration file: 1. Open your client settings file (such as claude_desktop_config.json or cline_mcp_settings.json). 2. Add the files-vectorstore entry under the mcpServers object: json { "mcpServers": { "files-vectorstore": { "command": "npx", "args": ["-y", "@lishenxydlgzs/simple-files-vectorstore"], "env": { "WATCH_DIRECTORIES": "/path/to/your/directories" } } } } 3. Optionally set CHUNK_SIZE, CHUNK_OVERLAP, or IGNORE_FILE in the env block. 4. Save the configuration and restart the client.

What you can do with Simple Files Vectorstore

• Querying local technical documentation and notes using natural language rather than exact keyword matches. - Indexing project folders to let LLMs automatically retrieve relevant source snippets during development. - Monitoring active local working directories to keep embedded context updated as files are modified. - Inspecting vector database indexing status and total processed document counts through the get_stats tool.

Key facts

• https://github.com/lishenxydlgzs/simple-files-vectorstore
• AI Memory & Context, Files, Documents & PDFs
• embeddings, local-files, rag, semantic search, vector-search

Part of MCP Servers

What can Simple Files Vectorstore do?

Simple Files Vectorstore continuously monitors specified local folders, splits documents into chunks, generates vector embeddings, and enables semantic search. It exposes MCP tools that let language models retrieve relevant text passages based on natural language queries, as well as check document indexing statistics.

How do I configure directories to watch?

You can define target folders by setting the WATCH_DIRECTORIES environment variable to a comma-separated list of paths in your MCP configuration. Alternatively, set WATCH_CONFIG_FILE to the path of a JSON file containing a watchList array with individual files or directories.

Which MCP clients work with Simple Files Vectorstore?

Any client supporting the Model Context Protocol over stdio can use this server. Supported environments include the Claude Desktop application and the VSCode Cline extension, both of which allow registering the npx package through their respective JSON settings files.

Can I exclude specific files from indexing?

Yes. You can supply the IGNORE_FILE environment variable containing the file path to a gitignore-style file. Simple Files Vectorstore will use the pattern rules in that file to ignore matching documents and directories during the indexing process.