DEV Community

Cover image for VegaAI - The Self Driving Agile Team
capscode
capscode

Posted on

VegaAI - The Self Driving Agile Team

#devchallenge #notionchallenge #mcp #ai

Notion MCP Challenge Submission 🧠

This is a submission for the Notion MCP Challenge

What I Built

VegaAI — The Self-Driving Agile Team (Powered by Notion MCP)

VegaAI is an AI-powered "Self-Driving Agile Team" platform where specialized agents— PM, Scrum Master, Dev, and QA—collaborate to plan and build software, one sprint at a time (Can be configurable to run multiple sprint, and multiple stories per sprint).

The core of VegaAI is Notion. Instead of just being a static documentation tool, Notion acts as the Executive Brain and State Machine for the entire AI workforce. Every PRD, feature, sprint, and story lives in Notion, and the agents use Notion MCP to read project state, update tasks, assign themselves to stories, and provide structured feedback per sprint.

Video Demo

Check out the walkthrough of the VegaAI

Show us the code

The repository for VegaAI is available on GitHub:

GitHub Repository — VegaAI

How I Used Notion MCP

Most AI coding tools use a local database. VegaAI is different. I’ve turned Notion into a real-time command center where the human and the AI can work on the same board simultaneously.

1. The "Agentic Toolbox" (MCP Methods Used)

I didn't just use one or two tools. I integrated the full power of the Notion MCP server across my entire agent fleet:

  • notion_create_page: Used by the PM Agent to document the initial PRD.
  • notion_create_database: Used once per project to build a customized Agile sprint board.
  • notion_create_database_item: Used by the Scrum Master to decompose features into actionable stories.
  • notion_update_page: The most used tool — utilized by ALL agents to flip statuses, update assignees, and append technical notes or comments.
  • notion_query_database: My UI's "Heartbeat" — used to ensure the Web Dashboard and Notion board stay in perfect sync.
  • notion_append_block_children: Scrum agent appends standup content to log page

2. Agent Workflow & Tool Specialization

My agents aren't just one big brain; they are specialists with their own focused toolsets:

  • PM/Scrum Agent: Orchestrates the "Agile Brain." It uses Notion MCP to set up the board and define Acceptance Criteria (AC) inside Notion properties.
  • Dev Agent: The "Builder." It reads the AC from Notion, then switches to the GitHub MCP to run github_create_repository and github_push_files and pushes the code to the repository.
  • QA Agent: The "Gatekeeper." It reviews the Dev's GitHub commits and then uses notion_update_page to record its result. If a AC fails, it writes a structured Failure: note back to Notion and re-assigns the story to the Dev agent.

3. Multi-LLM Orchestration

VegaAI is model-agnostic. I built a custom LLM Adapter that dynamically injects MCP tool schemas into the conversation. Whether you’re using OpenA, Anthropic, Groq or Gemini, the system "hot-loads" the Notion MCP tool definitions into the LLM's toolset during every turn.

4. The Robustness Loop (MAX_ATTEMPTS)

LLMs can occasionally hallucinate. I implemented a MAX_ATTEMPTS retry strategy. If the Dev agent fails or the QA agent finds a bug, the system resets the story status in Notion, QA mark the story as failed and assign it to DEV along with proper comments in the story and DEV tries again—preserving all previous failure notes so the next attempt is smarter. User can also move the story to next sprint manually if any story failed and I want to start new sprint.
Also after each dev work the DEV agent commit the code and push it to the GitHub repository via GitHub MCP.

5. Adding stories manually to the sprint

User can add stories manually to the sprint by clicking on the "Add Story" button in the UI. The story will be added to the current sprint and the agents will start working on it. User can also move the story to next sprint manually if any story failed and I want to start new sprint.

Why this matters

By integrating Notion MCP, I’ve moved "AI coding" from a black box CLI into a shared visual workspace. You can watch stories move from Todo → In Dev → In QA → Done in real-time as the agents work. It’s not just an AI tool; it’s a transparent, collaborative agency that lives in your Notion workspace.

What I will add in next version

  1. Multiple projects support
  2. Git branching per story + PRs
  3. Initiate new project.
  4. MCP servers for database and have full stack application support.
  5. WebSocket activity feed (replace polling)
  6. Persistent database
  7. More LLM support

Built with Next.js 14, Notion MCP, GitHub MCP, and the power of specialized agentic loops.

Top comments (0)