This is a submission for the Notion MCP Challenge
What I Built
NotionOps AI is an autonomous workflow engine that connects GitHub, OpenAI, and Notion into a single pipeline for software development teams.
It runs 4 specialized AI agents that automate the daily dev lifecycle:
| Agent | What it does |
|---|---|
| Issue Triage | Fetches open GitHub issues, classifies priority & type with AI, creates Notion tasks |
| PR Review | Reviews PR diffs, scores code quality 1-10, posts GitHub comments, creates Notion approval gates |
| Sprint Planner | Selects top tasks from backlog, estimates story points with AI, populates Notion sprint board |
| Daily Report | Aggregates sprint metrics, generates standup summary, saves report to Notion |
The agents chain together into a full pipeline:
GitHub Issues/PRs → OpenAI Agents → Human Review in Notion → Notion Databases
The dashboard provides real-time SSE streaming of agent activity, live stat cards, a kanban sprint board, and a full pipeline visualization.
Key design decision: The PR Review agent creates human-in-the-loop approval gates in Notion when it finds low-quality PRs (score < 6). AI never takes destructive action without human sign-off.
Screenshots
Dashboard
Pipeline Running
AI Agent Output
| Tasks View | Sprint Board | Reports |
|---|---|---|
 |
 |
 |
Notion Integration
| Tasks Database | Sprint Board | Agent Reports |
|---|---|---|
 |
 |
 |
Video Demo
Show us the code
stackmasteraliza
/
notionops-ai
Autonomous AI workflow engine — 4 Gemini-powered agents that triage GitHub issues, review PRs, plan sprints, and generate standups, all coordinated through Notion.
NotionOps AI — Software Dev Pipeline That Runs Itself 🤖
Notion MCP Challenge Submission — An autonomous AI workflow engine that closes the loop between GitHub, OpenAI, human judgment, and Notion — in real time.
What It Does
Most dev teams have a painful daily ritual: check GitHub for new issues, triage them manually, review PRs, update Notion, plan the sprint, write the standup. NotionOps AI eliminates every step of that with four specialized AI agents and a real-time neon dashboard.
GitHub Issues/PRs → OpenAI Agents → Human Review Gate → Notion (Tasks · Sprint · Reports)
↑ |
└───────────── feedback loop ───────────────┘
One click. The pipeline runs itself.
Demo Video
Four AI Agents
Agent
What it does
Output
🔍 Issue Triage
Fetches all open GitHub issues, classifies each as Bug/Feature/Issue with High/Medium/Low priority
Notion task per issue + triage report
👀 PR Review
Reviews every open PR, scores code
How I Used Notion MCP
Notion is the central coordination layer between AI agents and humans. Four databases drive the entire workflow:
1. Tasks Database
AI-triaged GitHub issues land here with structured metadata (priority, type, status, agent notes, GitHub URL). This is the single source of truth — the Sprint Planner reads from it, and humans can override AI classifications directly in Notion.
2. Approvals Database
When the PR Review agent finds a problematic PR, it creates an approval page with its recommendation. Team members review and approve/reject directly in Notion. This is the human-in-the-loop gate — downstream agents respect these decisions.
3. Sprint Database
The Sprint Planner reads pending tasks from Notion, uses AI to estimate complexity and assign Fibonacci story points, then writes selected tasks back with Backlog/In Progress/Done status tracking. The dashboard renders this as a live kanban board.
4. Reports Database
Daily standup summaries are generated from sprint data and stored as rich-text report pages. This creates an automatic paper trail of team velocity and AI-generated insights over time.
The key unlock: agents write structured data into Notion, humans make decisions in a familiar interface, and downstream agents read those decisions to continue the workflow. No context switching between GitHub, AI tools, and project management — it all flows through Notion.
Tech Stack
- Runtime: Node.js + TypeScript
- AI: OpenAI via OpenRouter (structured JSON outputs)
- APIs: Notion API, GitHub REST API (Octokit)
- Server: Express.js with Server-Sent Events
- Frontend: Vanilla HTML/CSS/JS (zero framework dependencies)
About Me
I'm Aliza Ali — a developer who loves building AI-powered tools that solve real workflow problems. If you liked this project, check out my other challenge submission:
FixForward: One Command to Go from Broken Build to Ready-to-Merge PR — A CLI autopilot that detects failing tests, classifies bugs, generates fixes via GitHub Copilot CLI, and creates verified PRs. One command. Three ecosystems. Zero manual patching.
Want to know more? Visit alizaali.com
Top comments (0)