Installation

Default Setup

TeaRAGs is pre-configured for MacBook M1/M2/M3 out of the box. Qdrant is built-in — it downloads and runs automatically as an embedded process. No Docker required.

No configuration required if you're running:

• Ollama on http://localhost:11434
• Default embedding model: unclemusclez/jina-embeddings-v2-base-code:latest

Just install, index, and search — defaults work immediately. Qdrant starts automatically on first use.

Prerequisites & Installation

Qdrant (Built-in)

Qdrant is embedded — TeaRAGs automatically downloads the Qdrant binary and manages it as a child process. No Docker, no manual setup.

How it works:

• On first use, TeaRAGs downloads the Qdrant binary for your platform
• Qdrant runs as a managed child process alongside TeaRAGs
• Data is stored in ~/.tea-rags/qdrant/storage by default
• Override storage location with QDRANT_EMBEDDED_STORAGE_PATH

Autodetect behavior (default, QDRANT_URL unset):

1. Probe localhost:6333 for an existing external Qdrant
2. If found, use it (seamless upgrade for existing Docker setups)
3. If not found, start embedded Qdrant automatically

Force embedded Qdrant:

export QDRANT_URL=embedded

---

Using External Qdrant (Optional)

If you prefer to run Qdrant separately (Docker, Qdrant Cloud, etc.), set QDRANT_URL explicitly:

export QDRANT_URL=http://localhost:6333

Docker setup

docker run -d \
  --name qdrant \
  -p 6333:6333 \
  -v $(pwd)/qdrant_storage:/qdrant/storage \
  qdrant/qdrant:latest

With custom memory limit:

docker run -d \
  --name qdrant \
  -p 6333:6333 \
  -v $(pwd)/qdrant_storage:/qdrant/storage \
  --memory=4g \
  qdrant/qdrant:latest

Native installation

macOS:

brew install qdrant

Linux:

wget https://github.com/qdrant/qdrant/releases/latest/download/qdrant-x86_64-unknown-linux-gnu.tar.gz
tar -xzf qdrant-x86_64-unknown-linux-gnu.tar.gz
./qdrant

Verify External Qdrant

curl http://localhost:6333/healthz
# Should return: "healthy"

---

Ollama Setup

Option 1: Native Installation (Recommended for GPU)

macOS (Apple Silicon)

Recommended: Ollama Desktop App

On Mac with Apple Silicon (M1/M2/M3/M4), install the Ollama desktop app instead of Homebrew. The desktop app automatically uses Metal GPU acceleration, which is significantly faster for embedding generation. Homebrew installs a CLI-only version that may not leverage GPU optimally.

# Recommended: download the .dmg from https://ollama.com/download/mac
# Drag to Applications → launch → GPU acceleration works automatically

# Alternative: Homebrew (CLI-only, less optimal GPU)
brew install ollama
ollama serve

Linux

# Install script
curl -fsSL https://ollama.com/install.sh | sh

# Start Ollama
ollama serve

Windows

# Download installer from https://ollama.com/download
# Run the installer, Ollama starts automatically

Option 2: Docker

CPU-only (all platforms)

docker run -d \
  --name ollama \
  -p 11434:11434 \
  -v ollama_models:/root/.ollama \
  ollama/ollama:latest

With GPU (Linux + NVIDIA)

docker run -d \
  --name ollama \
  --gpus all \
  -p 11434:11434 \
  -v ollama_models:/root/.ollama \
  ollama/ollama:latest

GPU Access in Docker

Docker GPU support requires NVIDIA Container Toolkit on Linux. On macOS, GPU acceleration is not available in Docker — use the native Ollama app for Metal GPU support.

Pull Embedding Model

# Default code-specialized model (768 dimensions)
ollama pull unclemusclez/jina-embeddings-v2-base-code:latest

# Alternative: nomic-embed-text (768 dimensions)
ollama pull nomic-embed-text:latest

# Alternative: mxbai-embed-large (1024 dimensions)
ollama pull mxbai-embed-large:latest

Verify Ollama

# Check Ollama is running
curl http://localhost:11434/api/version

# Test embedding
ollama run unclemusclez/jina-embeddings-v2-base-code:latest "test"

---

Installation Summary

Component	Recommended	Alternative
Qdrant	Built-in (embedded, zero setup)	Docker or native binary
Ollama	Native (for GPU)	Docker (CPU-only on Mac)

Recommended setup for MacBook:

• Qdrant: Built-in (automatic, no setup needed)
• Ollama: Native (GPU acceleration)

Recommended setup for Linux:

• Qdrant: Built-in (automatic)
• Ollama: Native or Docker with GPU (both support GPU)

All-in-Docker setup:

• Best for: Quick testing, CPU-only, remote servers
• Trade-off: No GPU acceleration on macOS

---

When to Configure

Configure environment variables when:

• Using remote Qdrant or remote GPU server for embeddings
• Switching to cloud providers (OpenAI, Cohere, Voyage AI)
• Performance tuning for large codebases (1M+ LOC)
• Enabling git enrichment (authorship, churn, bug-fix rates)

Essential Environment Variables

Set these in your MCP server configuration when deviating from defaults:

Connection URLs

Variable	Default	When to change
QDRANT_URL	Autodetect (probe localhost, fallback to embedded)	External Qdrant on remote server, or "embedded" to force built-in
EMBEDDING_BASE_URL	http://localhost:11434	Ollama on remote GPU server or custom port

Example: Remote GPU server with external Qdrant

claude mcp add tea-rags -s user -- node /path/to/tea-rags/build/index.js \
  -e QDRANT_URL=http://192.168.1.100:6333 \
  -e EMBEDDING_BASE_URL=http://192.168.1.100:11434

Embedding Provider

Variable	Default	Options
EMBEDDING_PROVIDER	ollama	ollama, openai, cohere, voyage
EMBEDDING_MODEL	unclemusclez/jina-embeddings-v2-base-code:latest	Provider-specific model name

Example: OpenAI

claude mcp add tea-rags -s user -- node /path/to/tea-rags/build/index.js \
  -e EMBEDDING_PROVIDER=openai \
  -e EMBEDDING_MODEL=text-embedding-3-small \
  -e OPENAI_API_KEY=sk-...

Git Enrichment

Variable	Default	Purpose
CODE_ENABLE_GIT_METADATA	false	Enable git blame analysis (authorship, churn, task IDs)

Enable trajectory enrichment:

claude mcp add tea-rags -s user -- node /path/to/tea-rags/build/index.js \
  -e CODE_ENABLE_GIT_METADATA=true

tip

Git enrichment runs concurrently with embedding and does not increase indexing time. See Git Enrichments for the full list of signals and advanced usage.

Performance Tuning

Variable	Default	When to change
EMBEDDING_BATCH_SIZE	1024	Tune via npm run tune for your hardware
INGEST_PIPELINE_CONCURRENCY	1	Increase for remote GPU (2–4 typical)
QDRANT_UPSERT_BATCH_SIZE	100	Tune via npm run tune

Auto-tune for your setup:

npm run tune

Generates tuned_environment_variables.env with optimal settings in ~60 seconds.

info

For detailed benchmarks, batch size optimization, and hardware-specific recommendations, see the Performance Tuning Guide.

Configuration Workflow

1. Start with defaults

# Install and run with zero configuration
npm install
npm run build
claude mcp add tea-rags -s user -- node /path/to/tea-rags/build/index.js

2. Configure only what you need

# Remote Qdrant + remote Ollama example
claude mcp add tea-rags -s user -- node /path/to/tea-rags/build/index.js \
  -e QDRANT_URL=http://gpu-server:6333 \
  -e EMBEDDING_BASE_URL=http://gpu-server:11434

3. Performance tune (optional)

# Auto-tune for your hardware (Qdrant auto-starts if not running)
EMBEDDING_BASE_URL=http://gpu-server:11434 \
npm run tune

# Apply tuned values
claude mcp add tea-rags -s user -- node /path/to/tea-rags/build/index.js \
  -e EMBEDDING_BASE_URL=http://gpu-server:11434 \
  -e EMBEDDING_BATCH_SIZE=256 \
  -e INGEST_PIPELINE_CONCURRENCY=4 \
  -e QDRANT_UPSERT_BATCH_SIZE=384

Quick Reference: Common Setups

Local MacBook (default)

# No configuration needed — defaults work out of the box
claude mcp add tea-rags -s user -- node /path/to/tea-rags/build/index.js

Remote GPU Server (with external Qdrant)

claude mcp add tea-rags -s user -- node /path/to/tea-rags/build/index.js \
  -e QDRANT_URL=http://192.168.1.100:6333 \
  -e EMBEDDING_BASE_URL=http://192.168.1.100:11434 \
  -e INGEST_PIPELINE_CONCURRENCY=4

OpenAI Embeddings

claude mcp add tea-rags -s user -- node /path/to/tea-rags/build/index.js \
  -e EMBEDDING_PROVIDER=openai \
  -e EMBEDDING_MODEL=text-embedding-3-small \
  -e OPENAI_API_KEY=sk-...

Production with Git Enrichment

claude mcp add tea-rags -s user -- node /path/to/tea-rags/build/index.js \
  -e CODE_ENABLE_GIT_METADATA=true

Next Steps

• Configuration Variables — full list of all configuration options
• Embedding Providers — detailed comparison of Ollama, OpenAI, Cohere, Voyage
• Performance Tuning — benchmarks and optimization guide