Agent Integration

Agent connections

Jira, Confluence and Slack reach your agent through the vendors' own MCP servers, running alongside the Elixium MCP.

Elixium holds no Atlassian or Slack credential and copies none of their content. Each developer's agent authenticates as you, reads what you can read, and any update it posts back carries your identity.

Setup

1. Add the Atlassian MCP and/or the Slack MCP to the same agent client that runs the Elixium MCP. Both authenticate as you; no admin step in Elixium.
2. Paste the full URL of a Jira issue, Confluence page or Slack thread into a story's text. Elixium lists it under External References when an agent orients on the story, and offers to mirror a transition to a linked Jira issue.

Bare issue keys such as PROJ-412 are ignored on purpose: Elixium's own readable IDs (INDTEK-812) share that shape and fill our story bodies, so a bare-key matcher would put phantom tickets in every brief. Paste the link.

Two limits to know

• Elixium cannot fence content it does not fetch. Your agent reads the vendor directly, so treat an unfamiliar page or thread as untrusted input; the boundary is your agent client's, not Elixium's.
• Elixium cannot see what those calls cost. Its usage reporting covers only its own contribution; take the vendor-side figure from your agent client.

One Integration, Every AI Agent

Elixium uses the Model Context Protocol (MCP) — the universal standard adopted by GitHub Copilot, Claude, Cursor, Gemini, and more. Configure once, use everywhere.

One Path Across Every Subscription

Every deployment model works the same way — your agent connects to an MCP endpoint. Only the endpoint (and how you authenticate to it) differs:

The per-agent instructions below show the hosted URL; self-hosted deployments substitute their own connect URL. The connector is deployed with Elixium itself, so an air-gapped install serves it inside your network and no traffic leaves it.

Claude (Desktop & Web): One-Click Connector

Connect Your Coding Agent

Every agent uses the same remote entry: one URL, OAuth sign-in on first use, no credential in any file. Self-hosted and air-gapped deployments use their own connect host in place of connect.elixium.ai.

claude mcp add --transport http elixium https://connect.elixium.ai/mcp

{
  "mcpServers": {
    "elixium": {
      "type": "http",
      "url": "https://connect.elixium.ai/mcp",
      "headers": { "X-Elixium-Board": "<YOUR_BOARD_SLUG>" }
    }
  }
}

{
  "servers": {
    "elixium": {
      "type": "http",
      "url": "https://connect.elixium.ai/mcp",
      "headers": { "X-Elixium-Board": "<YOUR_BOARD_SLUG>" }
    }
  }
}

{
  "mcpServers": {
    "elixium": {
      "type": "http",
      "url": "https://connect.elixium.ai/mcp",
      "headers": { "X-Elixium-Board": "<YOUR_BOARD_SLUG>" }
    }
  }
}

{
  "mcpServers": {
    "elixium": {
      "httpUrl": "https://connect.elixium.ai/mcp"
    }
  }
}

[mcp_servers.elixium]
url = "https://connect.elixium.ai/mcp"

{
  "mcpServers": {
    "elixium": {
      "serverUrl": "https://connect.elixium.ai/mcp"
    }
  }
}

Pin Your Board

A workspace can have many boards. Pinning tells every session which board this repo works against — unpinned agents fall back to your server-side default, which can silently point at the wrong board.

Prerequisites

Available MCP Tools

Once connected, your AI agent can use these tools:

Teaching Your AI Agent

Add a rules file to your repo so the AI knows how to use Elixium effectively.

# Elixium Integration

At the start of each session:
1. Call `get_iteration_context` to load the board
2. Review Current lane for active work

When implementing a story:
1. Call `prepare_implementation` for full context
2. Follow TDD: write tests first
3. Update story state to "started"
4. Commit with story ID in message

When complete:
1. Update story state to "finished"
2. Call `record_learning` with outcomes

Security

Troubleshooting