Local FAISS MCP Server

A Model Context Protocol (MCP) server that provides local vector database functionality using FAISS for Retrieval-Augmented Generation (RAG) applications.

Features

Core Capabilities

• Local Vector Storage: Uses FAISS for efficient similarity search without external dependencies
• Document Ingestion: Automatically chunks and embeds documents for storage
• Semantic Search: Query documents using natural language with sentence embeddings
• Persistent Storage: Indexes and metadata are saved to disk
• MCP Compatible: Works with any MCP-compatible AI agent or client

v0.2.0 Highlights

• CLI Tool: local-faiss command for standalone indexing and search
• Document Formats: Native PDF/TXT/MD support, DOCX/HTML/EPUB with pandoc
• Re-ranking: Two-stage retrieve and rerank for better results
• Custom Embeddings: Choose any Hugging Face embedding model
• MCP Prompts: Built-in prompts for answer extraction and summarization

Quickstart

# Install
pip install local-faiss-mcp

# Index documents
local-faiss index document.pdf

# Search
local-faiss search "What is this document about?"

Or use with Claude Code - configure MCP client (see Configuration) and try:

Use the ingest_document tool with: ./path/to/document.pdf
Then use query_rag_store to search for: "How does FAISS perform similarity search?"

Claude will retrieve relevant document chunks from your vector store and use them to answer your question.

Installation

⚡️ Upgrading? Run pip install --upgrade local-faiss-mcp

From PyPI (Recommended)

pip install local-faiss-mcp

Optional: Extended Format Support

For DOCX, HTML, EPUB, and 40+ additional formats, install pandoc:

# macOS
brew install pandoc

# Linux
sudo apt install pandoc

# Or download from: https://pandoc.org/installing.html

Note: PDF, TXT, and MD work without pandoc.

From Source

git clone https://github.com/nonatofabio/local_faiss_mcp.git
cd local_faiss_mcp
pip install -e .

Usage

Running the Server

After installation, you can run the server in three ways:

1. Using the installed command (easiest):

local-faiss-mcp --index-dir /path/to/index/directory

2. As a Python module:

python -m local_faiss_mcp --index-dir /path/to/index/directory

3. For development/testing:

python local_faiss_mcp/server.py --index-dir /path/to/index/directory

Command-line Arguments:

• --index-dir: Directory to store FAISS index and metadata files (default: current directory)
• --embed: Hugging Face embedding model name (default: all-MiniLM-L6-v2)
• --rerank: Enable re-ranking with specified cross-encoder model (default: BAAI/bge-reranker-base)

Using a Custom Embedding Model:

# Use a larger, more accurate model
local-faiss-mcp --index-dir ./.vector_store --embed all-mpnet-base-v2

# Use a multilingual model
local-faiss-mcp --index-dir ./.vector_store --embed paraphrase-multilingual-MiniLM-L12-v2

# Use any Hugging Face sentence-transformers model
local-faiss-mcp --index-dir ./.vector_store --embed sentence-transformers/model-name

Using Re-ranking for Better Results:

Re-ranking uses a cross-encoder model to reorder FAISS results for improved relevance. This two-stage "retrieve and rerank" approach is common in production search systems.

# Enable re-ranking with default model (BAAI/bge-reranker-base)
local-faiss-mcp --index-dir ./.vector_store --rerank

# Use a specific re-ranking model
local-faiss-mcp --index-dir ./.vector_store --rerank cross-encoder/ms-marco-MiniLM-L-6-v2

# Combine custom embedding and re-ranking
local-faiss-mcp --index-dir ./.vector_store --embed all-mpnet-base-v2 --rerank BAAI/bge-reranker-base

How Re-ranking Works:

1. FAISS retrieves top candidates (10x more than requested)
2. Cross-encoder scores each candidate against the query
3. Results are re-sorted by relevance score
4. Top-k most relevant results are returned

Popular re-ranking models:

• BAAI/bge-reranker-base - Good balance (default)
• cross-encoder/ms-marco-MiniLM-L-6-v2 - Fast and efficient
• cross-encoder/ms-marco-TinyBERT-L-2-v2 - Very fast, smaller model

The server will:

• Create the index directory if it doesn't exist
• Load existing FAISS index from {index-dir}/faiss.index (or create a new one)
• Load document metadata from {index-dir}/metadata.json (or create new)
• Listen for MCP tool calls via stdin/stdout

Available Tools

The server provides two tools for document management:

1. ingest_document

Ingest a document into the vector store.

Parameters:

• document (required): Text content OR file path to ingest
• source (optional): Identifier for the document source (default: "unknown")

Auto-detection: If document looks like a file path, it will be automatically parsed.

Supported formats:

• Native: TXT, MD, PDF
• With pandoc: DOCX, ODT, HTML, RTF, EPUB, and 40+ formats

Examples:

{
  "document": "FAISS is a library for efficient similarity search...",
  "source": "faiss_docs.txt"
}

{
  "document": "./documents/research_paper.pdf"
}

2. query_rag_store

Query the vector store for relevant document chunks.

Parameters:

• query (required): The search query text
• top_k (optional): Number of results to return (default: 3)

Example:

{
  "query": "How does FAISS perform similarity search?",
  "top_k": 5
}

Available Prompts

The server provides MCP prompts to help extract answers and summarize information from retrieved documents:

1. extract-answer

Extract the most relevant answer from retrieved document chunks with proper citations.

Arguments:

• query (required): The original user query or question
• chunks (required): Retrieved document chunks as JSON array with fields: text, source, distance

Use Case: After querying the RAG store, use this prompt to get a well-formatted answer that cites sources and explains relevance.

Example workflow in Claude:

1. Use query_rag_store tool to retrieve relevant chunks
2. Use extract-answer prompt with the query and results
3. Get a comprehensive answer with citations

2. summarize-documents

Create a focused summary from multiple document chunks.

Arguments:

• topic (required): The topic or theme to summarize
• chunks (required): Document chunks to summarize as JSON array
• max_length (optional): Maximum summary length in words (default: 200)

Use Case: Synthesize information from multiple retrieved documents into a concise summary.

Example Usage:

In Claude Code, after retrieving documents with query_rag_store, you can use the prompts like:

Use the extract-answer prompt with:
- query: "What is FAISS?"
- chunks: [the JSON results from query_rag_store]

The prompts will guide the LLM to provide structured, citation-backed answers based on your vector store data.

Command-Line Interface

The local-faiss CLI provides standalone document indexing and search capabilities.

Index Command

Index documents from the command line:

# Index single file
local-faiss index document.pdf

# Index multiple files
local-faiss index doc1.pdf doc2.txt doc3.md

# Index all files in folder
local-faiss index documents/

# Index recursively
local-faiss index -r documents/

# Index with glob pattern
local-faiss index "docs/**/*.pdf"

Configuration: The CLI automatically uses MCP configuration from:

1. ./.mcp.json (local/project-specific)
2. ~/.claude/.mcp.json (Claude Code config)
3. ~/.mcp.json (fallback)

If no config exists, creates ./.mcp.json with default settings (./.vector_store).

Supported formats:

• Native: TXT, MD, PDF (always available)
• With pandoc: DOCX, ODT, HTML, RTF, EPUB, etc.
  • Install: brew install pandoc (macOS) or apt install pandoc (Linux)

Search Command

Search the indexed documents:

# Basic search
local-faiss search "What is FAISS?"

# Get more results
local-faiss search -k 5 "similarity search algorithms"

Results show:

• Source file path
• FAISS distance score
• Re-rank score (if enabled in MCP config)
• Text preview (first 300 characters)

CLI Features

• ✅ Incremental indexing: Adds to existing index, doesn't overwrite
• ✅ Progress output: Shows indexing progress for each file
• ✅ Shared config: Uses same settings as MCP server
• ✅ Auto-detection: Supports glob patterns and recursive folders
• ✅ Format support: Handles PDF, TXT, MD natively; DOCX+ with pandoc

Configuration with MCP Clients

Claude Code

Add this server to your Claude Code MCP configuration (.mcp.json):

User-wide configuration (~/.claude/.mcp.json):

{
  "mcpServers": {
    "local-faiss-mcp": {
      "command": "local-faiss-mcp"
    }
  }
}

With custom index directory:

{
  "mcpServers": {
    "local-faiss-mcp": {
      "command": "local-faiss-mcp",
      "args": [
        "--index-dir",
        "/home/user/vector_indexes/my_project"
      ]
    }
  }
}