longbatman09/r2p-enterprise

Purpose and Entry Points

R2P Enterprise is a multi-tenant school platform that converts unstructured student reports into extracted marks, charts, and optional presentation decks, while offering isolated per-student textbook RAG. Its production boundary is the FastAPI MCP backend: mcp_backend/main.py, with dashboard REST endpoints and authenticated MCP JSON-RPC/SSE access. Schools primarily consume it through r2p_school_sdk/client.py.

API, Identity, and Tenant Boundary

The backend supports Supabase email/password sessions for dashboard users and sk_... API keys for SDK/MCP callers. JWT/API-key authentication is unified at the MCP boundary; service credentials can be stored for users, with AES-GCM protection when configured. School, student, key, invoice, and report entities are scoped to the school tenant.

• mcp_backend/auth_routes.py

• mcp_backend/auth_supabase.py

• mcp_backend/school_routes.py

• mcp_backend/database.py

• supabase/schema.sql

Report Analysis Pipeline

A caller uploads a report through an MCP tool. The UI orchestration layer coordinates prescan/upload analysis; vision extraction invokes Gemini to identify exam, student, subject, marks, and attendance from PDFs or images. Local pipeline agents merge extracted file/page results into student-oriented JSON, then chart/PPT rendering produces user-facing artifacts.

• mcp_servers/ui_orchestrator.py

• mcp_servers/vision_extractor.py

• agents/orchestrator.py

• agents/local_mem.py

• mcp_servers/plot_renderer.py

Retrieval-Augmented Textbook Service

Textbook ingestion and question answering are separate from report extraction. Content is embedded using NVIDIA NIM and stored/searchable in Pinecone namespaces partitioned by student, preventing cross-student retrieval. The MCP RAG module is the service-side implementation; the SDK exposes the school-facing call shape.

• mcp_servers/rag_system.py

• r2p_school_sdk/rag/manager.py

• r2p_school_sdk/rag/pinecone_helper.py

• migrate_rag_supabase.sql

School SDK and Commercial Integrations

r2p_school_sdk is the thin Python integration layer: it sends authenticated report and RAG requests to the backend rather than exposing MCP mechanics to schools. It also contains administrative/dashboard access and Stripe invoicing support. Backend persistence remains authoritative for tenant data and API-key lifecycle.

• r2p_school_sdk/client.py

• r2p_school_sdk/admin/dashboard.py

• r2p_school_sdk/invoicing/stripe_service.py

Runtime and Deployment

The primary runtime is Python/FastAPI/Uvicorn in Docker, deployed via Render. Supabase provides Auth and Postgres; Google Gemini provides document extraction; NVIDIA NIM and Pinecone provide RAG; Stripe handles invoices. Local composition and Render provisioning define the backend deployment path.

• mcp_backend/Dockerfile

• mcp_backend/requirements.txt

• docker-compose.yml

• render.yaml

Secondary Agent Runtimes

Two Google Agents CLI/ADK applications appear to be independently deployable agent runtimes, likely for conversational/orchestration and scanning roles. They are separate from the primary FastAPI MCP service but include telemetry and Terraform deployment definitions.

• my-agent/app/agent_runtime_app.py

• scan-agent/app/agent_runtime_app.py

• my-agent/deployment/terraform/single-project/service.tf