Skip to content

Installation

Four paths to a running DepthFusion instance. Pick the one that matches your environment.

Option A — pip (local, no GPU)

The default for developer laptops and CI.

pip install depthfusion
python -m depthfusion.install.install

Starts a local process-per-session server. No REST API, no Docker required. MCP config:

{
  "mcpServers": {
    "depthfusion": {
      "command": "python",
      "args": ["-m", "depthfusion.mcp.server"]
    }
  }
}

Option B — Docker (local or VPS)

docker pull ghcr.io/digittal-ai/depthfusion:latest

docker run -d \
  --name depthfusion \
  -p 127.0.0.1:7300:7300 \
  -v ~/.config/depthfusion:/data \
  -e DEPTHFUSION_REST_API=true \
  ghcr.io/digittal-ai/depthfusion:latest

Then point your MCP config at the HTTP server:

{
  "mcpServers": {
    "depthfusion": {
      "command": "python",
      "args": ["-m", "depthfusion.mcp.server"],
      "env": { "DEPTHFUSION_REST_API": "http://127.0.0.1:7300" }
    }
  }
}

Option C — VPS CPU

For self-hosted VPS deployments without a GPU. BM25 retrieval, no embedding model required.

→ VPS CPU Quickstart

Option D — macOS MLX

For Apple Silicon Macs. Runs local embedding inference via MLX.

→ macOS MLX Quickstart

After installation