Skip to content

docker.md

Latest commit

 

History

History
109 lines (76 loc) · 3.73 KB

docker.md

File metadata and controls

109 lines (76 loc) · 3.73 KB
Raw

Docker

Vera provides Docker images for running the MCP server (or any Vera command) in a container. Four variants are available:

Image Base Size Use case
vera:cpu debian:trixie-slim ~150 MB Default, works everywhere
vera:cuda nvidia/cuda:13.1.1-runtime ~4 GB NVIDIA GPU acceleration
vera:rocm rocm/dev-ubuntu-24.04:6.4.4 ~8 GB AMD GPU acceleration
vera:openvino ubuntu:24.04 + OpenVINO ~1 GB Intel GPU/iGPU acceleration

Running the MCP server

The primary use case is running Vera's MCP server as a background process for your editor or AI agent.

CPU:

docker run --rm -i -v $(pwd):/workspace ghcr.io/veratools/vera:cpu

CUDA (NVIDIA):

docker run --rm --gpus all -i -v $(pwd):/workspace ghcr.io/veratools/vera:cuda

ROCm (AMD):

docker run --rm --device=/dev/kfd --device=/dev/dri -i -v $(pwd):/workspace ghcr.io/veratools/vera:rocm

OpenVINO (Intel):

docker run --rm --device=/dev/dri -i -v $(pwd):/workspace ghcr.io/veratools/vera:openvino

The container starts vera mcp by default (JSON-RPC over stdio). The -i flag keeps stdin open for communication. The volume mount gives Vera access to your project files.

MCP client configuration

Point your MCP client (Claude Desktop, Cursor, etc.) at the Docker container:

{
  "mcpServers": {
    "vera": {
      "command": "docker",
      "args": ["run", "--rm", "-i", "-v", "/path/to/project:/workspace", "ghcr.io/veratools/vera:cpu"]
    }
  }
}

For GPU variants, add the appropriate device flags to args:

{
  "args": ["run", "--rm", "--gpus", "all", "-i", "-v", "/path/to/project:/workspace", "ghcr.io/veratools/vera:cuda"]
}

First run

On first run inside the container, Vera will download ONNX models and the ONNX Runtime library to /root/.vera/. To persist these across container restarts, mount a volume:

docker run --rm -i \
  -v $(pwd):/workspace \
  -v vera-models:/root/.vera \
  ghcr.io/veratools/vera:cpu

After the first run, subsequent starts skip the download.

Running other commands

Override the default mcp command to run any Vera command:

docker run --rm -v $(pwd):/workspace ghcr.io/veratools/vera:cpu index /workspace
docker run --rm -v $(pwd):/workspace ghcr.io/veratools/vera:cpu search "authentication logic"
docker run --rm -v $(pwd):/workspace ghcr.io/veratools/vera:cpu stats

Building locally

The Dockerfiles package a prebuilt binary, so build it first. The binary must be a Linux build matching the image architecture (the published images are linux/amd64), so run the build on a Linux x86_64 machine, or on macOS/Windows inside a Linux container, e.g. docker run --rm -v $(pwd):/src -w /src rust:latest bash -c './scripts/bootstrap-vendored-grammars.sh && cargo build --release -p vera-cli'. From the repo root:

./scripts/bootstrap-vendored-grammars.sh  # downloads grammars required by vera-core's build script
cargo build --release -p vera-cli
mkdir -p dist && cp target/release/vera dist/

docker build -f docker/Dockerfile.cpu -t vera:cpu .
docker build -f docker/Dockerfile.cuda -t vera:cuda .
docker build -f docker/Dockerfile.rocm -t vera:rocm .
docker build -f docker/Dockerfile.openvino -t vera:openvino .

GPU requirements

CUDA: Requires NVIDIA drivers with CUDA 12+ support and the NVIDIA Container Toolkit.

ROCm: Requires AMD GPU with ROCm 6.x drivers. The /dev/kfd and /dev/dri devices must be accessible.

OpenVINO: Requires Intel GPU/iGPU with the Intel compute runtime installed. The /dev/dri device must be accessible. Linux x86_64 only.