Agent orchestration playbooks

External LLMs (Claude, ChatGPT, Cursor, Grok, Copilot) should treat https://fifa.qtech.hr/mcp as the FIFA Agent Mentor intelligence layer: fewer, higher-signal tool calls with structured JSON + markdown reports.

Auth: Authorization: Bearer <fam_live_… | Supabase access token>.

Agent Pro unlocks market + Cases tools. Exam Academy unlocks study AI (ask_study_assistant unlimited, generate_mock_exam, drill_weak_topic).

Why this matters

A “pure FIFA agent controlled by AI” does not invent regulations or valuations. It calls tools, reads structured results, cites sources, and writes decisions back into Cases / dossiers when needed.

Job	Prefer
Full transfer picture	orchestrate_transfer_dossier (one call)
Weak-topic exam prep	drill_weak_topic then lesson / mock links
Jurisprudence case file	Cases sequence below
Live market briefing	get_market_pulse_report or generate_market_pulse_report

1. Full transfer picture (one call)

orchestrate_transfer_dossier

• Args: player_slug, optional days_back (default 21), include_market_pulse: true, live_synthesis: true for one xAI synthesis pass (extra cost).

Returns: timeline subset, agent_memories, agent_dossiers, optional Durable Object snapshot, optional market pulse excerpt, optional Grok synthesis.

2. Classic multi-step research

1. research_player (slug_or_name, depth: deep)
2. dossier_snapshot
3. get_transfer_intel
4. get_market_pulse_report (period: daily | weekly)

3. Market pulse

• get_market_pulse_report — read cron output (daily 07:25 UTC, weekly Mon 07:40 UTC).
• generate_market_pulse_report — on-demand Grok + DB pipeline (writes a row; optional webhook).

4. Watch list

monitor_player needs a user-bound token. Combine with watched_players for ongoing monitoring.

5. Agent Law Cases — end-to-end

1. list_cases({ status: 'open' })
2. get_case({ case_id })
3. build_case_source_pack({ case_id, depth?, query? })
4. list_case_sources / get_case_findings
5. attach_case_source / add_case_link / add_case_note (notes are not auto-indexed in v1)
6. search_case({ case_id, question, ... }) — stateless
7. request_case_research({ case_id, query, depth, seed_urls?, use_browser? }) — quick (Grok) | deep/exhaustive (+ Browser Run/Kitesurf)
8. search_juris_library({ q, forum?, year? }) — always cite source_url

Citation rules: never invent FIFA articles or decision IDs; never re-host full FIFA decision PDFs; if retrieval is weak, say so.

6. Exam study loop (Study Tutor Agent)

1. list_lessons → get_lesson
2. ask_study_assistant (Bearer) with lesson_number when focused
3. Paid: drill_weak_topic — retrieve → adaptive MCQs → answer key → platform links
   (also POST https://fifa.qtech.hr/api/study/drill)
4. Paid: generate_mock_exam for full timed sets

When the student says “drill / quiz me / weak topic”, prefer drill_weak_topic. Non-premium → PREMIUM_REQUIRED + /pricing.

Example — weak-topic drill

{
  "jsonrpc": "2.0",
  "id": 2,
  "method": "tools/call",
  "params": {
    "name": "drill_weak_topic",
    "arguments": {
      "topic": "RSTP Article 19 protection of minors",
      "lesson_number": 4,
      "question_count": 3
    }
  }
}

7. Operator automation

• Mint keys: POST /api/internal/mcp/mint with x-worker-secret (see Ops).
• Health: GET /api/health.

---

Related: MCP overview · Premium tools · Exam tools · Remote connectors · app skill fifa-agent-orchestration.md