DEV Community

Yash Sonawane
Yash Sonawane

Posted on

NotionOps AI — I Built an AI DevOps Brain That Lives Inside Notion 🧠

#devchallenge #notionchallenge #mcp #ai

Notion MCP Challenge Submission 🧠

This is a submission for the Notion MCP Challenge

What I Built

Every developer knows this pain:

It’s 11 PM. A production bug hits.

You jump between GitHub → Jira → Slack → Notion → Terminal…
Five tools. Zero clarity.

This constant context switching is a hidden tax on every engineering team.

So I built NotionOps AI — an AI-powered DevOps brain that lives inside Notion.

👉 It automatically connects:

  • GitHub activity
  • AI analysis (Claude)
  • Notion workspace

And turns them into a single source of truth.

What It Does

  • Every commit, PR, and issue → automatically becomes a task
  • Critical issues → instantly become incidents
  • Deployments → logged without manual work
  • Daily standups → generated automatically

Notion becomes your team’s central nervous system. AI does the rest.

Live Demo Flow

  1. Developer pushes: hotfix: null pointer in payment gateway
  2. GitHub webhook triggers
  3. AI detects CRITICAL severity
  4. A P0 incident page is created in Notion
  5. A high-priority task is generated
  6. AI standup updates automatically next morning

💡 Result:
No manual updates. No missed incidents. No confusion.

Show us the code

👉 GitHub Repository:
https://github.com/yashsonawane25/NotionOps-AI

Repo includes:

  • MCP server (10 tools)
  • FastAPI webhook system
  • AI analysis engine
  • Notion integration layer
  • Local simulation script

How I Used Notion MCP

Notion MCP is the core backbone of this system — not just an add-on.

I designed a 4-database architecture inside Notion:

1. Tasks Database

  • Auto-created from GitHub events
  • Includes priority, category, and status
  • No manual input required

2. Deployments Database

  • Tracks every deployment
  • Stores environment, version, and status

3. Incidents Database

  • Auto-created for critical issues
  • Includes severity (P0/P1), description, and logs

4. AI Digest Database

  • Daily standups generated automatically
  • AI reads tasks → writes summary

MCP Tools (Core Innovation)

I exposed 10 tools via MCP, including:

  • analyze_commit
  • analyze_pr
  • create_task
  • log_incident
  • log_deployment
  • generate_standup
  • query_project
  • process_github_push

🔥 Killer Feature: process_github_push

One function does everything:

➡️ Analyze commit
➡️ Create task
➡️ Detect severity
➡️ Trigger incident (if critical)

🧠 AI Query Feature (Game Changer)

You can literally ask:

“What are our highest priority tasks?”

And get a real-time answer from Notion.

No dashboards. No filters. Just ask.

Architecture 

GitHub Webhook
      │
      ▼
FastAPI Server
      │
      ▼
Claude AI Analysis
      │
      ▼
MCP Server
      │
      ▼
Notion API
Enter fullscreen mode Exit fullscreen mode

Tech Stack

  • Python
  • FastAPI
  • MCP SDK
  • Claude API (Anthropic)
  • Notion API
  • Pydantic

Running It Yourself 

git clone https://github.com/yashsonawane25/NotionOps-AI
cd NotionOps-AI
pip install -r requirements.txt
cp .env.example .env

# Add your API keys

uvicorn webhook_server:app --reload --port 8000

# Run simulation
python test_simulate.py
Enter fullscreen mode Exit fullscreen mode

What I Learned

  • MCP forces clean, modular design
  • Async processing is critical for webhooks
  • Notion works surprisingly well as a DevOps backend
  • Simulation scripts save massive dev time

What’s Next

  • Bidirectional sync (Notion → GitHub)
  • AI-based velocity tracking
  • Weekly executive reports
  • On-call automation

Final Thought

I didn’t want to build a demo.

I built something I would actually use in a real DevOps team.

NotionOps AI = GitHub + AI + Notion → Fully automated DevOps workflow

The best DevOps system is the one that works… even when you don’t.

Credits

Built by Yash Sonawane
B.Tech — GH Raisoni College of Engineering & Management, Pune

GitHub: https://github.com/yashsonawane25

Tags

notion mcp devops ai python hackathon automation github claude

Top comments (3)

Collapse
 
klement_gunndu profile image
klement Gunndu •

Smart move using Claude to classify commit severity — how do you handle false positives where a routine refactor triggers a P1 incident?

Collapse
 
yash_sonawane25 profile image
Yash Sonawane •

Great question honestly!

So yeah, false positives are a real problem. My current fix is pretty simple — if the commit message has tags like [refactor] or [chore], the system just skips the critical path entirely. Also, Claude gives back a confidence score along with the severity, so anything below a certain threshold doesn't auto-create an incident, it goes into a "needs review" state instead.

There's also a small delay before the incident page actually gets created, so if something looks wrong you can catch it in time.

But honestly it's not perfect yet. The next version will have a feedback loop where devs can flag false positives, and that'll help improve the AI prompts over time.

What would you do differently? Curious to hear your approach!

Some comments may only be visible to logged-in visitors. Sign in to view all comments.