Guide

Knowledge / RAG / Sub-MCP

Anamaya MCP exposes OMS tools only. It does not embed a RAG engine. Policy composition runs in your LLM host or orchestrator.

Employee → Client chat → Client LLM / orchestrator
                              │
                              ├─(A) Client knowledge: RAG / DB / HR / Sub-MCP / files
                              │
                              └─(B) Anamaya MCP → OMS → inventory

Mode A — RAG over your database / APIs

Resolve employee context (grade, cost centre, caps, approver) before search. Filter offers in code for money-risk fields, then call Anamaya MCP.

{
  "employee_email": "ada@acme.com",
  "oms_email": "ada@acme.com",
  "grade": "M3",
  "hotel_nightly_cap_idr": { "Jakarta": 1500000 },
  "banned_carriers": [],
  "require_surat_jalan": true
}

Mode B — Customer Sub-MCP

Register both servers in the host: anamaya + your policy MCP. Call your tools first (policy_get_employee, policy_get_rules, policy_check_offer), then Anamaya search/book tools.

Mode C — Static knowledge packs

Versioned JSON/MD for slow-changing rules. Prefer structured JSON for hard caps; keep prose for explanations only.

Per-turn sequence

1. Identify employee → oms_email
2. Load guardrails (A/B/C) + optional anamaya_travel_policies
3. Search live via Anamaya MCP
4. Filter / rank ≤3
5. User confirms
6. Book saga with confirm: true
7. Poll + anamaya_files_fetch