RAG Memory & Model Context Protocol (MCP)

Cora transforms static studio data into a living intelligence layer via Retrieval-Augmented Generation (RAG) and open Model Context Protocol (MCP) server endpoints.

1. RAG Memory Architecture

Unlike complex external vector database clusters that introduce latency, Cora uses a self-contained keyword-dense semantic chunking engine stored in MySQL/SQLite:

1. Ingestion: Uploaded PDFs, rate cards, and client history are split into 512-token chunks.
2. Indexing: Chunks are indexed with semantic entity tags in wp_cora_rag_knowledge.
3. Retrieval: When an AI prompt is executed, the top 5 most relevant studio context chunks are automatically injected into the LLM system prompt.

2. MCP JSON-RPC 2.0 Gateway

Cora acts as an MCP Server, allowing external desktop IDEs (Cursor, Windsurf, Claude Desktop, Antigravity) to query workspace data directly over WebSockets and REST:

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "cora_get_shoot_schedule",
    "arguments": {
      "date": "2026-08-30"
    }
  },
  "id": 1
}

Available MCP Tools

• cora_query_leads: Search active CRM deals and stage milestones.
• cora_create_invoice: Generate 18% GST invoice PDF.
• cora_check_gear_availability: Query inventory catalog for camera/lens conflicts.
• cora_dispatch_call_sheet: Send WhatsApp notification to crew members.