Elixium Logo
elixium.aiAgile | XP

IDE Integration

Connect Elixium directly to your development workflow. Access your backlog, update story status, and get AI guidance without leaving your IDE.

Elixium MCP Server (Antigravity)

Elixium exposes a Model Context Protocol (MCP) server that allows AI agents (like Cursor, Windsurf, or Antigravity) to read and manipulate your project board.

If you do not have the Elixium codebase, install the MCP server from npm instead of building from source.

# Install the MCP server (optional)

npm install -D @elixium.ai/mcp-server@latest

# Or install globally

npm install -g @elixium.ai/mcp-server@latest

Security & Multi-Tenancy

Elixium uses a secure, tenant-scoped API key system. Each key is hardcoded to your specific workspace in the database. This ensures that even if a key is compromised, it only provides access to your specific projects and data.

1. Generate an API Key

Request an administrative API key from your workspace owner. They can provide you with a key from the sys_api_keys collection.

2. Configure your Agent

Update your MCP configuration. CRITICAL: Use your tenant subdomain in the ELIXIUM_API_URL to ensure the server resolves the correct context.

In VS Code, put this config in .vscode/mcp.json.

To target a specific board within your workspace, set ELIXIUM_BOARD_SLUG. The MCP server resolves the slug to a boardId at runtime, so the slug must match an existing board.

If your API stores lanes in uppercase (e.g. CURRENT/BACKLOG), set ELIXIUM_LANE_STYLE to upper.

{
  "mcpServers": {
    "elixium": {
      "command": "npx",
      "args": [
        "-y",
        "@elixium.ai/mcp-server@latest"
      ],
      "env": {
        "ELIXIUM_API_KEY": "<YOUR_API_KEY>",
        "ELIXIUM_API_URL": "https://<YOUR_TENANT>.elixium.ai/api",
        "ELIXIUM_BOARD_SLUG": "main"
      }
    }
  }
}

If you built the MCP server from source, set command to node and args to the absolute path of dist/index.js.

Tip: To restart the MCP server without reloading VS Code, update and save .vscode/mcp.json (even a whitespace change). Most MCP clients watch the file and restart the server automatically.

VS Code Extension

The official Elixium VS Code extension provides a dedicated sidebar for MCP tools and iteration context, with strict guardrails.

  • Browse MCP tools and iteration context (Current, Backlog, Objectives).
  • Start story flow and copy an implementation brief.
  • Create stories and record learnings from the Command Palette.
  • Guardrails: the extension blocks moving stories into Current/Done and only allows state changes to started.
  • Open stories in the board and create git branches from story titles.
Why guardrails exist: Federal and air-gapped teams require strict separation between automated assistance and human commitment. These guardrails keep the extension in a read/assist role while preserving auditability and decision authority.
📦 Install from Marketplace(Coming Soon)

IDE Agent Guidelines & Process

Elixium is designed to be "Agent-First." When using an AI agent (like Antigravity) in your IDE, follow the **Pivotal Way** methodology to ensure high-quality, hypothesis-driven development.

The Elixium Philosophy

  • Small Slices: Break work into the smallest testable units.
  • TDD First: No implementation without a failing test.
  • Learning Loop: Treat every story as a hypothesis to validate.

The Rules File

Antigravity/Windsurf uses antigravity/elixium.rules for shared agent guidance. Other IDEs have their own conventions:

  • Cursor: Use .cursorrules (legacy) or .cursor/rules/*.md (current). Docs
  • VS Code + GitHub Copilot: Use .github/copilot-instructions.md if you enable repo-level instructions, or document guidance in your workspace settings. Docs
  • Codex CLI: Add AGENTS.md at the repo root for shared instructions when running Codex. Docs

Standard Workflow: /implement-story

When starting a new task, use the /implement-story slash command. The agent will follow this automated process:

  1. Identify: The agent lists stories in your Current lane.
  2. Prepare: The agent runs the prepare_implementation tool to fetch AC, learning goals, and perform guardrail checks.
  3. Planning Brief: The agent presents an Implementation Brief. You must approve the Acceptance Criteria, Assumptions, and the Proposal before any code is written.