AI in Aztra
Two ways to use AI — built-in AI with your own API key, or connect your AI assistant through MCP.
Aztra's AI works with your actual projects, goals, and progress — so answers and summaries are grounded in real data, not guesses. There are two ways to use it, and they're independent: use either, or both.
| AI with your key | AI through MCP | |
|---|---|---|
| What it is | Built-in AI features inside Aztra | Your own AI assistant connected to Aztra |
| Powered by | Your OpenAI API key | Whatever AI tool you already use |
| Where it runs | In the Aztra UI | In your assistant (Claude, and other MCP-capable tools) |
| Set up by | Workspace owner or admin | Each user, from their own AI tool |
AI with your key (built-in)
Bring your own OpenAI API key and Aztra's in-app AI features light up:
- AI chats — conversations scoped to your workspace, with saved history. Available to owners, admins, and managers.
- AI-assisted reports — from the Reports area, generate a written summary of what shipped over a period.
- AI test cases — on an issue's QA tab, generate suggested test cases from the issue's own description.
Open AI settings
Go to Settings → AI (owner or admin). AI stays off until you set it up.
Add your API key and enable AI
Paste your OpenAI API key and choose the model. The key is stored encrypted and used only for your workspace.
Use it across the product
AI chats, report generation, and test-case suggestions all use these settings.
AI through MCP (bring your assistant)
Aztra runs an MCP server — the open Model Context Protocol — so an AI assistant you already use can work with your workspace directly: reading and updating issues, sprints, goals, wiki pages and time logs, acting as you and limited to what you can already see and do.
It works with Claude.ai, Claude Desktop, Claude Code, ChatGPT, Cursor, VS Code and any other MCP client, either by signing in from the client or with a key you mint in Aztra.
Connect your AI assistant → — setup for each client, what your assistant can do, limits, and how to revoke access.
The two modes complement each other: your key powers AI inside Aztra for the whole workspace; MCP brings Aztra into the AI tools each person already works in.