directory-indexer by peteretelej - MCP Server

Directory Indexer

Turn your directories into an AI-powered knowledge base.

Self-hosted semantic search for local files. Give your AI assistant access to search, read, and work with your local files and directories. Ask about topics, patterns, or examples and get meaningful results from your actual knowledge base.

Quick Start

Prerequisites:

Docker - For running Qdrant and Ollama (skip if you already have them running natively)
Node.js 18+ - Required for running directory-indexer

Note: For native Qdrant and Ollama installation without Docker, see Setup section.

1. Start Qdrant vector database (skip if already running)

docker run -d --name qdrant -p 127.0.0.1:6333:6333 -v qdrant_storage:/qdrant/storage qdrant/qdrant

2. Start Ollama embedding service (skip if already running)

Note: Docker Ollama won't use GPU acceleration. For better performance, consider native installation.

docker run -d --name ollama -p 127.0.0.1:11434:11434 -v ollama:/root/.ollama ollama/ollama

# Pull the embedding model
docker exec ollama ollama pull nomic-embed-text

Note: Make sure the embedding model is pulled before you start indexing.

You can confirm that Ollama and Qdrant are running by checking:

# qdrant endpoint
curl http://localhost:6333/

# confirm what ollama models are available
curl http://localhost:11434/api/tags

3. Index your directories

npx directory-indexer@latest index ./WorkNotes ./Projects

4. Configure AI assistant (Claude Desktop, Cursor, Cline, Roo Code, Zed etc.)

Add to your MCP configuration:

{
  "mcpServers": {
    "directory-indexer": {
      "command": "npx",
      "args": ["directory-indexer@latest", "serve"]
    }
  }
}

If you experience issues on windows adding this MCP. You can install the package globally using

npm install -g directory-indexer@latest

Then use the following MCP configuration

{
  "mcpServers": {
    "directory-indexer": {
      "command": "directory-indexer",
      "args": [ "serve" ]
    }
  }
}

Your AI assistant will automatically start the MCP server and can now search your indexed files.

Advanced: For organizing content into focused search areas, see Workspace Support. For faster indexing, see Performance Tips.

Setup

Directory Indexer runs locally on your machine or server. It uses an embedding provider (such as Ollama) to create vector embeddings of your files and stores them in a Qdrant vector database for fast semantic search. Both services can run remotely if needed.

Setup requires two services:

1. Qdrant Vector Database

Choose one option:

Docker (recommended for most users):

docker run -d --name qdrant \
    -p 127.0.0.1:6333:6333 \
    -v qdrant_storage:/qdrant/storage \
    qdrant/qdrant

This option requires Docker
Runs Qdrant on docker container, uses a named volume qdrant_storage for persistent storage.

Alternative: Install natively from qdrant.tech

2. Embedding Provider

Choose one option:

Option A: Ollama (recommended - free, runs locally)

# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh  # Linux/macOS
# For Windows: Download from https://ollama.ai

# Pull the embedding model
ollama pull nomic-embed-text

You can also run Ollama via Docker
- GPU support may require additional configuration

Option B: OpenAI (requires paid API key)

export OPENAI_API_KEY="your-api-key-here"

Quick Verification

Test your setup:

# Check Qdrant Health
curl http://localhost:6333/healthz

# View collections
curl http://localhost:6333/collections

# Check Ollama
curl http://localhost:11434/api/tags

If either fails, directory-indexer will show a helpful error with setup guidance.

Usage

MCP Integration

Configure with AI assistants (Claude Desktop, Cline, etc.) using npx:

{
  "mcpServers": {
    "directory-indexer": {
      "command": "npx",
      "args": ["directory-indexer@latest", "serve"]
    }
  }
}

Index your directories:

# Index your directories first
npx directory-indexer@latest index /home/user/projects/docs /home/user/work/reports

How it works:

MCP server starts automatically - When your AI assistant connects, it launches the MCP server in the background
Indexing runs independently - You can index files before, during, or after MCP setup
Search immediately available - Your AI assistant can search files as soon as they're indexed

Key point: You don't need to wait for indexing to complete before using the MCP server. Index files as needed, and your AI assistant will immediately have access to search them.

Using with AI Assistant

Once configured, your AI assistant can search your indexed documents semantically:

Search by concept:

"Find API authentication examples"
"Show me error handling patterns"
"Find configuration for Redis"

Find similar content:

"Show me incidents similar to this outage report" (when you have an incident file open)
"Find documentation like this API guide" (when viewing an API doc)
"What files are similar to my deployment script?"

Troubleshoot issues:

"Find troubleshooting guides on SQL deadlocks"
"Show me solutions for timeout errors"
"Find debugging tips for performance issues"

Custom Configuration

Configure with custom endpoints and data directory:

{
  "mcpServers": {
    "directory-indexer": {
      "command": "npx",
      "args": ["directory-indexer@latest", "serve"],
      "env": {
        "DIRECTORY_INDEXER_DATA_DIR": "/opt/ai-knowledge-base",
        "QDRANT_ENDPOINT": "http://localhost:6333",
        "OLLAMA_ENDPOINT": "http://localhost:11434"
      }
    }
  }
}

Workspace Support

Organize content into workspaces for focused searches:

{
  "mcpServers": {
    "directory-indexer": {
      "command": "npx",
      "args": ["directory-indexer@latest", "serve"],
      "env": {
        "WORKSPACE_CUSTOMER_CASES": "C:\\Users\\john\\Documents\\Support\\Cases,C:\\Users\\john\\Documents\\Incidents",
        "WORKSPACE_ENGINEERING_DOCS": "C:\\Users\\john\\Code\\API,C:\\Users\\john\\Code\\Web",
        "WORKSPACE_COMPANY_POLICIES": "C:\\Users\\john\\Documents\\Policies,C:\\Users\\john\\Documents\\Procedures"
      }
    }
  }
}

How workspaces work:

Define workspace environments with WORKSPACE_NAME format
Use comma-separated paths or JSON arrays: ["path1", "path2"]
Search within specific workspaces: "Find issues about authentication in customer cases workspace"
Your AI assistant can filter results to relevant workspace content
Use server_info to see available workspaces and their statistics

CLI Usage

For advanced users who prefer command-line usage, see .

Configuration

Environment variables (all optional):

# Data directory (default: ~/.directory-indexer)
export DIRECTORY_INDEXER_DATA_DIR="/opt/ai-knowledge-base"

# Service endpoints (defaults shown)
export QDRANT_ENDPOINT="http://localhost:6333"
export OLLAMA_ENDPOINT="http://localhost:11434"

# Optional API keys
export OPENAI_API_KEY="your-key-here"
export QDRANT_API_KEY="your-key-here"

For all configuration options, see .

Performance Tips

Speed up embedding generation:

Install Ollama natively - Enables automatic GPU acceleration (Docker version uses CPU only)
Use OpenAI API - Faster than local embeddings but requires paid API key and sends data to OpenAI servers

Smart indexing:

Index parent directories - Avoids duplicates since full file paths are stored
Keep folders focused - Only index directories with documents you want searchable
Start with key folders - Index your most important documentation first

Time management:

Run during off-hours - Let large directories index overnight
Continue working - MCP server works immediately while indexing runs in background
Re-index efficiently - Only changed files are reprocessed when you re-run indexing

Supported Files

Text: .md, .txt
Code: .rs, .py, .js, .ts, .go, .java, etc.
Data: .json, .yaml, .csv, .toml
Config: .env, .conf, .ini
Upcoming: Support for more file types like PDFs, docx, etc, see #11

Documentation

: CLI commands and MCP tools
: System architecture and process flows
: Development setup and guidelines
: Architecture and technical decisions

License

MIT