OmniPlan

Initial UI
Basic chats -which the AI contextualises itself on (location, sentient analysis etc)
Ability to switch between different themes
Ability to add files, reply etc
Calling for the AI to execute
Output : after processing through 6 MCP servers and Claude, shows the optimal restaurant, along with custom directions for each user

💡 Inspiration

We’ve all been there: trying to plan a simple dinner with friends turns into a chaotic nightmare. Sarah is vegan, Tom has allergies, nobody remembers who can’t do Tuesdays, and someone inevitably suggests a place 2 hours away during rush hour.

We realized the real problem isn't that we can't chat; it's that we are making decisions blind. AI should be able to help, but standard chatbots can’t "read minds" - they lack context. They don't know where you are right now, what the weather is like, or that you hate spicy food.

We wanted to solve this by building OmniPlan: a tool that combines every user's calendar, location, preferences, and real-time dynamic context in one place.

🤖 What it does

OmniPlan is a hyper-personalized group chat assistant powered by the Model Context Protocol (MCP). It acts as a silent observer that only chimes in when you tag @AI.

When activated, OmniPlan doesn't just "guess" a recommendation. It spins up 5 distinct MCP servers to gather live context: Checks Calendars: Finds the true free time for everyone. Sentiment Analysis: Reads the chat history to understand the group's mood and preferences. Profile Matching: Updates and references individual dietary restrictions and budget constraints. Real-Time Location: Checks locations to find a central meeting point (so nobody travels 2 hours). Live Weather: Checks if it’s raining to avoid outdoor seating suggestions.

The result is a smart suggestion that matches all criteria, complete with custom directions for every single person in the chat.

⚙️ How we built it

We built OmniPlan as a modular agent leveraging the Model Context Protocol (MCP) to standardize how the AI accesses information.

The Brain: We used Claude as the orchestration engine because of its ability to work with live, hyper-personalized data streams. The Architecture: Instead of a monolithic backend, we built five separate MCP servers (calendar, weather, location, directions, and sentiment) that act as "tools" for the LLM. Modular Design: We used a modular MCP design so new context sources (like weather or events) could be added instantly. Privacy: The bot uses a specific workflow that ensures only defined tools can run, helping to prevent hallucinations and protect group chat privacy.

🚧 Challenges we ran into

The MCP Learning Curve: Since MCP is a relatively new standard, simply getting the workflow to ensure only our defined tools ran took significant effort.

Context Overload: Balancing the amount of data fed to the context window was tricky. We had to ensure the sentiment analysis understood both individual and group preferences in context without overwhelming the model.

🏆 Accomplishments that we're proud of

True Modularity: We successfully implemented a modular MCP design where new context sources can be added instantly.

Privacy-First Design: We built the bot to be passive. It only responds when @AI is called, which protects group-chat privacy. Working Prototype: We delivered a fully working, real-time contextual assistant using multiple MCP servers with a clean, intuitive UX.

🧠 What we learned

The Power of Context: We learned that AI needs more than just a prompt; it needs the "deep context" of calendars, location, and weather to be truly useful. Real-Time Constraints: Working with live context streaming required us to optimize how the model handles hyper-personalized data.

🚀 What's next for OmniPlan

We plan to evolve OmniPlan from a meeting scheduler into a full travel and social companion. Our roadmap includes: Richer API Integrations: Integrating Google Maps/Transit, restaurant availability, live events, and flight status. Proactive Alerts: Adding proactive alerts for weather shifts, transport delays, or nearby food matches. Adaptability: Making the assistant adapt to where you are and what is happening around you in real-time.