🧠 MCP Context Engine
A minimal context-aware middleware that bridges real-world data (like your calendar) to AI systems — enabling intelligent, time-aware responses and decisions.
🚀 Inspiration
I wanted to explore how AI agents can reason with real-world signals like user schedules, meetings, and time zones — similar to how the Model Context Protocol (MCP) aims to unify AI context access.
The goal was to build something small but production-grade, showing how a model could answer:
“Am I free tomorrow morning for a 45-minute meeting?”
This led to the creation of a lightweight context provider API that reads .ics calendar data, interprets natural-language queries, and returns structured JSON context — ready for AI reasoning.
🧩 How I Built It
The system was designed with a layered FastAPI architecture, emphasizing modularity and extensibility.
⚙️ Core Design
- FastAPI for high-performance async API endpoints
- httpx for external data fetching
- icalendar for robust
.icsparsing - pydantic-settings for environment configuration
- dateutil / tzdata for accurate timezone conversion
🧱 Milestones
| Milestone | Description |
|---|---|
| M0 | FastAPI scaffold with config, logging, and error handling |
| M1 | Health & diagnostics routes |
| M2 | Google Calendar (ICS) provider with event parsing and diagnostics |
| M3 | Natural-language availability + slot suggestion engine ✅ |
🧠 What I Learned
- How MCP-like context protocols structure metadata for LLMs
- Parsing and normalizing ICS calendar data for time-based reasoning
- Handling time zone consistency in scheduling logic
- Designing natural language interfaces to structured time data
- Writing clean modular FastAPI services that scale
💡 Challenges Faced
- Parsing and interpreting recurring events from
.icsdata - Aligning human time expressions (like “tomorrow morning”) with concrete UTC timestamps
- Managing time zone drift between calendar and query input
- Building a natural-language slot finder that respects event overlaps and buffer windows
🧮 Example Queries
| Query | Output |
|---|---|
| “Am I free tomorrow at 03:10?” | → {"busy": true, "event": "Daily Sync"} |
| “Any slot tomorrow morning for 45 min?” | → {"start": "09:30", "end": "10:15"} |
🧰 Tech Stack
- Backend: FastAPI, Uvicorn
- Parsing: icalendar, dateutil
- Networking: httpx
- Environment: pydantic-settings, dotenv
- Testing: pytest, curl / Swagger
📸 Screenshots
🧭 Swagger Home
📅 Busy Check
⏰ Slot Suggestion
🧾 Example: Slot Suggestion Logic (LaTeX)
The slot engine computes free intervals by taking the complement of all busy periods within a given time window:
[ \text{Free Slots} = \text{All Slots} \setminus \text{Busy Intervals} ]
For a 45-minute duration ( d ), the algorithm searches:
[ \exists\, t \in [t_\text{start}, t_\text{end}] \;\text{s.t.}\; (t, t + d) \cap \text{Busy} = \emptyset ]
🌟 Summary
The MCP Context Engine demonstrates how AI systems can connect with live human context — like schedules and time — through a simple yet powerful middleware layer.
It’s a step toward contextually intelligent AI, capable of reasoning not just with data, but with your world.
