How to Connect AI Agents to Stellary via MCP

The Model Context Protocol (MCP) is the bridge between your AI agents and your Stellary workspace. In this guide, we'll walk through the complete setup — from generating your API token to having your first agent propose actions on your board.

What is MCP?

MCP is an open protocol that lets AI models interact with external tools and data sources in a standardized way. Think of it as a universal adapter: any MCP-compatible agent can connect to any MCP-compatible tool without custom integration code.

Stellary implements MCP natively, which means your workspace becomes a rich context source that agents can read from and act upon. For a deeper dive into what MCP is and why it matters, see MCP Explained.

Prerequisites

Before you begin, make sure you have:

• A Stellary workspace with at least one project
• An API token (generated from Settings > API Tokens)
• An MCP-compatible AI agent (Claude, GPT, or any custom agent)

Step 1: Generate Your API Token

Navigate to your workspace settings and create a new API token. Give it a descriptive name — you'll want to know which agent is using which token.

# You can also create tokens via the API
curl -X POST https://api.stellary.co/v1/api-tokens \
  -H "Authorization: Bearer YOUR_SESSION_TOKEN" \
  -d '{"name": "My Claude Agent", "scopes": ["read", "write"]}'

The response will include your token. Store it securely — it won't be shown again.

Step 2: Configure Your Agent

Point your agent's MCP configuration to your Stellary workspace. The exact setup depends on your agent, but the core configuration looks like this:

{
  "mcpServers": {
    "stellary": {
      "url": "https://api.stellary.co/v1/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_TOKEN"
      }
    }
  }
}

Step 3: Verify the Connection

Once configured, your agent should be able to list available tools. You'll see tools for reading your board, creating cards, updating priorities, and more.

Available MCP Tools

Stellary exposes a comprehensive set of tools through MCP:

• Board tools — list columns, create/update/move cards, manage labels
• Doc tools — read and create documents, trigger review workflows
• Cockpit tools — read missions, log decisions, propose actions
• Project tools — list projects, read settings, manage members

Best Practices

Here are some tips for getting the most out of MCP integration:

• Start with read-only scopes — let your agent observe before it acts
• Use the cockpit for proposals — agents should propose actions, not execute them directly
• Monitor agent activity — check the automation traces to see what your agents are doing
• Set up approval workflows — require human approval for destructive actions

What's Next

Now that your agent is connected, explore our guide on setting up automations to let your agent work autonomously within safe boundaries.