hpcGPT is a customized CLI built on top of the Opencode agent that integrates Model Context Protocol (MCP) servers for Slurm-based HPC environments, Illinois Chat documentation Q&A, reporting, and Atlassian.

TL;DR – Getting Started

curl -fsSL https://opencode.ai/install | bash
export OPENCODE_CONFIG=/absolute/path/to/this/repo/opencode.jsonc
opencode

Set environment variables as needed (see Env section below), then pick a model and use tools from the TUI.

Features

• Slurm integration (MCP): sinfo, squeue, scontrol, and accounts via slurm-mcp-server.
• Docs Q&A (MCP): Illinois Chat tools delta-docs, delta-ai-docs for Delta/Delta AI documentation.
• Atlassian integration (MCP): Containerized Jira/Confluence tools with flexible auth modes.
• Support reporting: report-server sends concise support reports with context.
• Multiple providers: NCSA Hosted and NCSA Ollama providers selectable in opencode.jsonc.
• Config-driven: Everything wired through opencode.jsonc for reproducibility.

System Architecture

graph TD
  U[User] -->|TUI| OC[Opencode Agent]

  OC --> P1[NCSA Hosted Provider]
  OC --> P2[NCSA Ollama Provider]

  subgraph MCP_Servers
    M1[slurm-mcp-server]
    M2[illinois-chat-server]
    M3[report-server]
    M4[atlassian-mcp-server container]
  end

  OC -. tools .-> M1
  OC -. tools .-> M2
  OC -. tools .-> M3
  OC -. tools .-> M4

  M1 --> SLURM[Slurm CLI]
  M2 --> ICHAT[Illinois Chat API]
  M4 --> JIRA[Jira]
  M4 --> CONF[Confluence]
  M3 --> SUPPORT[Delta Support]

Loading

How things fit together

• Opencode reads opencode.jsonc for providers, models, and MCP servers.
• MCP servers expose tools over stdio; the agent calls them when the model chooses a tool.
• slurm-mcp-server shells out to local Slurm commands.
• illinois-chat-server calls the Illinois Chat API to answer questions from Delta/Delta AI docs.
• atlassian-mcp-server runs via Apptainer and exposes Jira/Confluence tools.
• report-server can send a compact support report with session context.

Project Structure

hpcgpt/
  mcp_servers/
    illinois_chat_server/
      src/index.ts
      package.json
    slurm_mcp_server/
      src/index.ts
      package.json
  prompts/
    support.txt
  opencode.jsonc
  example.env
  example.env.atlassian
  README.md
  favicon.png

MCP Servers & Tools

• slurm-mcp (local)

  • Tools: accounts, sinfo, squeue, scontrol
  • Purpose: query accounts, node/partition status, user jobs, and job details.

• illinois-chat-mcp (local)

  • Tools: delta-docs, delta-ai-docs
  • Purpose: answer questions from Delta and Delta AI documentation (requires ILLINOIS_CHAT_API_KEY).

• report-server (local)

  • Tools: send_support_report
  • Purpose: email a concise support report with conversation history and system info to the Delta support team.

• atlassian-mcp-server (container)

  • Tools (examples): Jira — jira_get_issue, jira_search_issues, jira_create_issue, jira_add_comment, jira_transition_issue; Confluence — confluence_search, confluence_get_page, confluence_create_page, confluence_update_page (availability depends on config/read-only mode).
  • Purpose: interact with Jira and Confluence for tickets and docs. See the Atlassian MCP project for details: https://github.com/sooperset/mcp-atlassian

Installation

Install Opencode and point it at this repo’s config:

curl -fsSL https://opencode.ai/install | bash
export OPENCODE_CONFIG=/absolute/path/to/this/repo/opencode.jsonc
opencode

Optional: Local MCP server setup

MCP servers in mcp_servers/* use Bun/Node. From each server directory:

bun install
bun run build
bun run start

Opencode will also launch them automatically from the opencode.jsonc mcp section when enabled.

Environment Configuration

Use example.env and example.env.atlassian as references. Export directly or create a .env/.env.atlassian.

Core variables

• NCSA_LLM_URL – Base URL for NCSA Hosted models provider
• ILLINOIS_CHAT_API_KEY – Required for illinois-chat-server

Atlassian (containerized MCP)

Configure .env.atlassian (see example.env.atlassian). Common options:

• JIRA_URL, CONFLUENCE_URL
• One of: personal token, API token, or OAuth BYOT
• Optional: READ_ONLY_MODE, ENABLED_TOOLS, proxy settings

Usage Examples

Inside the Opencode TUI, pick a model (e.g., ncsahosted/Qwen/Qwen3-VL-32B-Instruct) and ask the assistant to use tools.

Slurm status

"Check the Delta GPU partitions and my running jobs."

The assistant will call sinfo and squeue via slurm-mcp-server.

Delta/Delta AI docs Q&A

"How do I submit a Slurm job on Delta?"

The assistant will call delta-docs with your question and return a synthesized answer with citations when available.

File a support report

Run the report command in Opencode. This uses the send_support_report tool to email a concise summary to Delta support.

Configuration Reference

See opencode.jsonc for providers, models, and MCP server commands. Example provider entries:

{
  "provider": {
    "ncsahosted": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "my_provider_name",
      "options": {
        "baseURL": "{env:my_url}" // load the url from a environment variable
      },
      "models": {
        "Qwen/Qwen3-VL-32B-Instruct": {
          "name": "my_model_name",
          "options": {
            "stream":true
          }
        }
      }
    }
  }
}

Links

• Delta Chatbot: https://uiuc.chat/Delta-Documentation (course: Delta-Documentation)
• Delta AI Chatbot: https://uiuc.chat/DeltaAI-Documentation (course: DeltaAI-Documentation)

License

MIT – see LICENSE.