How to Connect AI Agents to Stellary via MCP
A practical guide to integrating AI agents with your Stellary workspace using the Model Context Protocol.
Mar 28, 20263 min read
You've probably tried asking ChatGPT for help with your project. It gives generic advice because it has no idea what your project actually looks like. That's the context problem — and it's the single biggest barrier to useful AI in project management.
The Context Problem
Generic AI Is Useless AI
Ask an AI "what should we prioritize?" without context and you get textbook answers: "Focus on high-impact items." "Address blockers first." "Consider your deadlines." This is what you'd find in any project management 101 article.
Now give that same AI access to your board, your sprint history, your team's velocity data, and your documented priorities. Suddenly it can say: "Card SP-47 has been in progress for 8 days — 3x your average. The assignee is also working on 4 other cards. Consider reassigning or descoping."
That's the difference context makes.
What Context Means in Practice
Context for an AI project assistant includes:
- Board state — what's in each column, who's assigned, what's blocked
- History — how the board has changed over time, what was completed, what was carried over
- Documents — architecture decisions, technical specs, meeting notes
- Priorities — what the team has explicitly declared as important
- Decisions — what choices were made and why
- Team structure — who has what skills, who's available, who's overloaded
Without this context, AI can only react to what you tell it in a chat window. With it, AI can proactively identify issues, suggest improvements, and propose actions.
Levels of AI Context
Level 0: No Context (Chatbot)
The AI sees only what you type. It can answer generic questions but has no knowledge of your specific project. This is where most teams are today.
Level 1: Snapshot Context
The AI can see the current state of your board — what's in each column right now. It can answer "what's blocked?" but can't say "this card has been blocked for an unusually long time."
Level 2: Historical Context
The AI can see how your project has evolved over time. It understands velocity trends, recurring patterns, and can compare the current sprint to past ones.
Level 3: Deep Context
The AI can access your board, your documents, your decisions, your priorities, and your team structure. It understands not just what's happening, but why. This is where AI becomes genuinely transformative.
How to Give AI the Right Context
Structured Data Over Raw Text
AI works better with structured data than with long text descriptions. A card with typed fields (priority: high, status: blocked, assignee: Alice, due: April 5) is more useful than a paragraph describing the same information.
Decision Logs
Every decision you log with context — the reasoning, the alternatives considered — becomes training data for AI recommendations. When a similar decision arises, the AI can surface what you decided last time and why.
Living Documentation
Documentation that lives next to your work and stays updated gives AI a rich knowledge base. Stale docs are worse than no docs — they actively mislead the AI.
MCP: The Context Protocol
The Model Context Protocol is specifically designed to solve the context problem. It gives AI agents structured access to your project data through a standardized interface. Instead of scraping or manual copy-paste, the AI gets clean, typed, real-time access to everything it needs.
Context-Aware AI in Action
Smarter Suggestions
Without context: "Consider adding more detail to your cards." With context: "SP-23 and SP-31 are both assigned to Alice and due this week. Based on her velocity, she'll likely complete only one. Consider reassigning SP-31."
Better Decisions
Without context: "Weigh the pros and cons of each option." With context: "Last quarter, you chose Redis over Memcached for caching (Decision #14). The reasoning was multi-data-type support. This new caching requirement has the same pattern — Redis is likely the right choice again."
Proactive Alerts
Without context: the AI is silent. With context: "3 cards in the 'In Review' column have been there for more than 5 days. Your average review cycle is 2 days. This may indicate a reviewer bottleneck."
The Competitive Advantage
Teams that give their AI tools deep context will dramatically outperform those that use AI as a glorified chatbot. The context gap is the difference between AI that generates busywork and AI that generates insight.
The investment is straightforward: use a tool that structures your project data well, keep your documentation alive, log your decisions, and connect via MCP. The AI does the rest.