overview index

Project overview

360 Ghar is a unified real estate platform backend built on FastAPI, PostgreSQL with PostGIS, SQLAlchemy 2.x async, and Pydantic v2. It runs six integrated modules from a single async codebase, authenticates against Supabase Auth, and exposes both a REST surface at /api/v1/* and two MCP servers at /mcp and /mcp-admin for LLM clients. The codebase is roughly 68,000 lines of Python across 352 source files, with 333 REST endpoints and 40+ MCP tools.

The six modules

Module	Purpose	Key paths
Ghar Core	Buy/rent marketplace with swipe-based discovery, geospatial search, agent coordination, and visits.	app/api/api_v1/endpoints/properties.py, swipes.py, visits.py, agents.py
360 Stays	Short-stay bookings for hotels, vacation rentals, and temporary stays with availability checks and dynamic pricing.	bookings.py, app/services/booking.py
360 Flatmates	Roommate and PG discovery with swipe matching, conversations, moderation, QnA, and visit scheduling.	app/services/flatmates/
Property Management	Landlord and PM tools: leases, tenants, rent collection, maintenance, documents, inspections, reports.	app/services/pm_*.py, endpoints pm_*.py
360 Virtual Tours	Immersive 360 degree tour builder with scenes, hotspots, floor plans, AI jobs, and analytics.	app/services/tour/, app/services/tour_ai/
360 Data Hub	Real estate data aggregation: bank auctions, RERA projects/complaints, circle rates, gazette, jamabandi, zoning, neighbourhood.	app/services/data_hub/

Who uses it

• Property seekers discover homes through swipe feeds and semantic search, schedule visits, and chat with agents.
• Owners and landlords list properties, manage leases and rent, track maintenance, and view portfolio dashboards.
• Agents coordinate visits, manage assignments, and run portfolio operations via the admin MCP server.
• LLM clients (ChatGPT, Claude Desktop, Cursor, VS Code Copilot) call MCP tools and render interactive widgets.
• Platform admins moderate flatmate listings, manage notifications, and run data hub scrapers.

Tech stack

Python 3.10+ with FastAPI, SQLAlchemy 2.x async, Pydantic v2, httpx, APScheduler, pgvector, GeoAlchemy2, Supabase, FastMCP 3.0.1, Pydantic AI, Pillow, Cloudinary, and Sentry. Dependencies are managed with uv and locked in uv.lock. PostGIS powers geospatial queries, pgvector powers semantic search, and Redis backs the cache and pub/sub layers.

Key wiki pages

• Architecture — layered structure, request flow, MCP architecture with diagrams.
• Getting started — prerequisites, install, build, test, run commands, env config.
• Glossary — domain and technical vocabulary used across the codebase and docs.
• Patterns and conventions — async-first, service layer, shared httpx clients, ruff rules.

Other areas worth bookmarking once written: API authentication, MCP servers, infrastructure, the data model reference, and feature pages for each module.

Where to start reading code

• app/factory.py — thin composition root that builds the FastAPI app.
• app/infrastructure/ — lifespan, middleware, routing, MCP mounts, scheduler.
• app/api/api_v1/api.py — REST router composition.
• app/main.py — entrypoint, Sentry init, /health and /config probes.
• app/core/config.py — settings.
• app/models/enums.py — 50+ enums that shape every domain object.

The codebase has only two TODO/FIXME markers, so what you read is what ships. For the full operating contract, see CLAUDE.md and AGENTS.md at the repo root.