Skip to content

MESA VS Code

Repo: idss-mesa/vscode · Image: harbor.cyverse.org/vice/mesa-vscode:latest (GPU: :gpu) · In the portal: Applications → MESA Apps → MESA VS Code

Visual Studio Code in the browser (code-server), with the MESA AI coding agents and CyVerse Data Store tools added1. The Cline extension comes with the MESA MCP servers already configured, so you can use an agent from the editor's side bar as well as from the terminal.

What's inside

Category Tools
IDE code-server with the Python, Jupyter, vscode-icons, and Cline extensions
Science Miniconda and Mamba (/opt/conda)
Transfer Globus Connect Server 5.4
AI agents and MCP Claude Code, Codex, OpenCode, Antigravity, Claude Code Router; the irods, mesa, formation, and filesystem MCP servers, for the CLIs and for Cline — see AI agents in the MESA apps
CyVerse data GoCommands, iRODS configuration, S3/OSN mounts, AWS CLI
Developer tools GitHub CLI, Git Credential Manager, Go 1.25, Node.js 22

Start it

  1. In the MESA Portal, open Applications → MESA Apps.
  2. On MESA VS Code, click Instant Launch or Launch with Options. See Starting applications.
  3. VS Code opens in a new tab on your Data Store home folder.

You can also start it from the Discovery Environment by searching for MESA VS Code.

First steps

  1. Open a terminal: Terminal → New Terminal (or Ctrl+`).
  2. Run cyverse-login to give the tools and agents your CyVerse access.
  3. Optionally run aiverde-setup to connect AI Verde models.
  4. Start an agent: claude, codex, opencode, or agy, or open Cline in the side bar and choose a model provider.

See AI agents in the MESA apps. Keep your work in the Data Store folder VS Code opened; the rest of the container is deleted when the analysis ends.

GPU build

For GPU work, launch the GPU build (harbor.cyverse.org/vice/mesa-vscode:gpu). It adds:

Adds Details
CUDA The CUDA 12.5 toolkit (nvcc, cuda-gdb, compute-sanitizer, Nsight command-line tools)
PyTorch environment /opt/conda/envs/pytorch (Python 3.13): PyTorch 2.14 (CUDA 12.6), transformers, accelerate, datasets, peft, sentence-transformers, bitsandbytes, Lightning, timm, CuPy, and more. It is the default VS Code interpreter and a Jupyter kernel named PyTorch 2.14 (CUDA 12.6)
Local LLMs An Ollama server on the GPU — see Local models on a GPU; the Continue extension is set up for the local model
Extensions NVIDIA Nsight (CUDA debugging), clangd for C++/CUDA, CMake Tools
GPU tools mesa-gpu-check, nvtop, nvitop

Compile CUDA code for the A16 GPUs with, for example, nvcc -gencode arch=compute_86,code=sm_86 kernel.cu. When you install Python packages that depend on PyTorch, add --extra-index-url https://download.pytorch.org/whl/cu126 so pip keeps the CUDA 12.6 build.

Run it on your own computer

docker run --rm -p 8080:8080 -e IPLANT_USER=$USER harbor.cyverse.org/vice/mesa-vscode:latest

Open http://localhost:8080. Outside CyVerse there is no password, so publish the port only on your own machine. The image is built for linux/amd64.

  1. MESA VS Code README, https://github.com/idss-mesa/vscode. ↩

Machine-readable versions of this page: Markdown twin · raw source on GitHub · llms.txt · llms-full.txt (whole site). See For AI agents.