Skip to content

πŸ“‘ Ikhtisar Week 3 β€” Amplify With Multi-Agent Systems And MCP ​

Ringkasan Menyeluruh Minggu Ketiga 5 Hari Β· 32 Video Β· Self-Hosting, OAuth2, MCP, Context Engineering, Sub-agents, Capstone Multi-Agent + Voice CRM


πŸ—ΊοΈ Peta Pembelajaran Minggu Ini ​

Week 3: AMPLIFY
═══════════════

Day 1 ─── Self-Hosting & AI Lokal
β”‚         Docker, Ollama, OpenRouter, DeepSeek
β”‚
Day 2 ─── Integrasi Lanjutan
β”‚         OAuth2, Google Drive, Firecrawl, Structured Outputs
β”‚
Day 3 ─── MCP (Model Context Protocol)
β”‚         Host/Client/Server, Transport, Firecrawl MCP,
β”‚         Hunter.io MCP, Claude AI integration
β”‚
Day 4 ─── Context Engineering & Capstone Pt.1
β”‚         Structured Outputs vs Tools, Sub-agents,
β”‚         Lethal Trifecta, Pipedrive CRM, RevOps, SDR
β”‚
Day 5 ─── Capstone Finale
          BDM Agent, Account Executive, Deal Recording,
          Demo Booking, ElevenLabs Voice Agent, Grand Finale

πŸ“… Ringkasan Per Hari ​

Day 1 β€” Self-Hosting n8n dengan Docker & Ollama ​

TopikDetail
Cloud vs Self-hostedCloud = mudah, managed; Self-hosted = kontrol penuh, gratis, private
Docker DesktopInstall untuk Windows/Mac, container = unit portabel isolasi
Docker Run n8n-p 5678:5678, -v n8n_data, --add-host=host.docker.internal:host-gateway
Owner AccountSetup saat pertama akses localhost:5678
OpenRouter + DeepSeekAPI key gratis, model DeepSeek R1 untuk agent pertama self-hosted
OllamaAI 100% lokal, host.docker.internal:11434, model Mistral 3
OAuth LimitationGoogle OAuth2 membutuhkan domain publik β†’ kompleks di self-hosted

Konsep Kunci: Self-hosting memberikan kontrol penuh atas data dan zero cost untuk n8n, namun memerlukan penanganan infrastruktur sendiri. Docker menyederhanakan deployment secara drastis.


Day 2 β€” Integrasi Lanjutan: OAuth2, Firecrawl & Structured Outputs ​

TopikDetail
3 Tipe AuthHeader Auth, Query Auth, OAuth2 (paling kompleks, paling aman)
Google OAuth2GCP Console β†’ Project β†’ API β†’ Consent Screen β†’ Credentials β†’ Redirect URI
Google Drive TriggerMonitoring folder, download file by ID, extract PDF
FirecrawlWeb scraping sebagai API/MCP β€” scrape, crawl, map, extract
Structured OutputsTeknik KRUSIAL β€” force LLM output JSON schema tertentu
HTTP Request NodeNode generik untuk API apapun, termasuk Firecrawl REST

Konsep Kunci: Structured Outputs adalah game-changer β€” memaksa LLM menghasilkan output yang pasti sesuai schema. Ini adalah landasan untuk workflow production yang reliable.


Day 3 β€” MCP: Model Context Protocol ​

TopikDetail
Definisi MCP"USB-C for AI" β€” standar Anthropic untuk koneksi AI ↔ Tools
ArsitekturHost β†’ Client β†’ Server (three-tier)
Transportstdio (lokal), SSE (legacy), Streamable HTTP (masa depan)
n8n sebagai MCP ClientTool node yang connect ke MCP Server eksternal
n8n sebagai MCP ServerMCP Server Trigger β†’ expose n8n workflow ke AI app
Firecrawl MCPSSE transport, $vars untuk API key
Hunter.io MCPBearer Auth, email finder/verifier
Claude AI IntegrationCustom MCP Connector β†’ n8n MCP Server β†’ workflow execution

Konsep Kunci: MCP membawa standarisasi ke koneksi AI-tools. n8n bisa menjadi client (mengakses MCP servers) maupun server (mengexpose workflow ke Claude, dsb). Keterbatasan n8n: tidak support stdio (hanya SSE/Streamable HTTP).


Day 4 β€” Context Engineering, Sub-agents & Capstone Pt.1 ​

TopikDetail
Structured Outputs vs ToolsSO = boring tapi bulletproof; Tools = flexible tapi unpredictable
Context EngineeringPhil Schmidt (Google DeepMind): seni packing context window optimal
Sub-agentsIndependent testing, reusable, context-efficient β€” tapi batasi kreativitas
Lethal TrifectaData privat + Komunikasi eksternal + Konten tidak terpercaya = bahaya
Anti-patternsHuman trap, solution-focused, content-generation-as-goal
Pipedrive CRMFree trial + API access β†’ Leads, Deals, Organizations, People
RevOps Sub-agentStructured Outputs approach β†’ Create Org β†’ Person β†’ Lead (ID chaining)
SDR Sub-agentTools approach β†’ Pipedrive Get Many + Gmail Create Draft

Konsep Kunci: Context Engineering = definisi inti AI engineering modern. Sub-agents memecah masalah besar, tapi harus seimbang dengan kebutuhan otonomi. Lethal Trifecta adalah framework keamanan wajib untuk setiap deployment agentic AI.


Day 5 β€” Capstone Finale: Multi-Agent Sales + Voice CRM ​

TopikDetail
Sub-workflow ConversionSDR + RevOps β†’ trigger "When executed by another workflow" + Publish
BDM AgentGPT-5.2, 3 tools (Prospecting/RevOps/SDR), Google Drive PDF trigger
DebuggingInput field mix-up, boolean vs string IF, missing email field
Deal RecordingWebhook trigger, Pipedrive Get People + Create Deal, Structured Outputs
Demo BookingWebhook trigger, Google Calendar Availability + Create Event
ElevenLabs Voice AgentAccount Executive persona, 2 webhook tools, phone number connected
Grand FinalePhone call β†’ Deal in Pipedrive β†’ Demo in Calendar β†’ Push notifications

Konsep Kunci: Seluruh platform sales otomatis β€” dari PDF ICP drop sampai telepon balik β€” bisa dibangun dalam hitungan jam menggunakan n8n + ElevenLabs. Arsitektur final: 2 main agents, 5 sub-agents, 6 integrasi, 7 LLM calls.


πŸ”‘ Konsep Inti Week 3 ​

1. Self-Hosting & Infrastruktur ​

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚  DEPLOYMENT OPTIONS                                      β”‚
β”‚                                                          β”‚
β”‚  ☁️ Cloud (n8n.io)          🏠 Self-Hosted (Docker)     β”‚
β”‚  β”œβ”€β”€ Managed infrastructure  β”œβ”€β”€ Full data control       β”‚
β”‚  β”œβ”€β”€ Auto-updates            β”œβ”€β”€ Zero platform cost      β”‚
β”‚  β”œβ”€β”€ Built-in OAuth          β”œβ”€β”€ Private deployment      β”‚
β”‚  β”œβ”€β”€ Pay per execution       β”œβ”€β”€ Custom integrations     β”‚
β”‚  └── Easiest setup           └── Manual maintenance      β”‚
β”‚                                                          β”‚
β”‚  🧠 Local AI (Ollama)                                   β”‚
β”‚  β”œβ”€β”€ 100% offline capable                                β”‚
β”‚  β”œβ”€β”€ No API costs                                        β”‚
β”‚  β”œβ”€β”€ Models: Mistral, Llama, etc.                        β”‚
β”‚  └── Access: host.docker.internal:11434                  β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

2. MCP Ecosystem ​

        β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
        β”‚           MCP ARCHITECTURE                   β”‚
        β”‚                                              β”‚
        β”‚  HOST (Claude, n8n, IDE)                     β”‚
        β”‚    └── CLIENT (embedded in host)             β”‚
        β”‚          └── SERVER (Firecrawl, Hunter, n8n) β”‚
        β”‚                                              β”‚
        β”‚  TRANSPORTS:                                 β”‚
        β”‚  β”œβ”€β”€ stdio    β†’ lokal, cepat (not in n8n)   β”‚
        β”‚  β”œβ”€β”€ SSE      β†’ remote, legacy              β”‚
        β”‚  └── Streamable HTTP β†’ remote, masa depan   β”‚
        β”‚                                              β”‚
        β”‚  n8n ROLES:                                  β”‚
        β”‚  β”œβ”€β”€ πŸ”Œ MCP Client β†’ akses MCP servers      β”‚
        β”‚  └── πŸ–₯️ MCP Server β†’ expose ke Claude, dll  β”‚
        β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

3. Context Engineering & Keamanan ​

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚  CONTEXT ENGINEERING                                       β”‚
β”‚                                                            β”‚
β”‚  "Seni memilih token input yang tepat agar output LLM     β”‚
β”‚   konsisten dengan business goal."                         β”‚
β”‚                                                            β”‚
β”‚  β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”        β”‚
β”‚  β”‚ System  β”‚ β”‚  Chat   β”‚ β”‚  Tools  β”‚ β”‚  RAG    β”‚        β”‚
β”‚  β”‚ Prompt  β”‚ β”‚ History β”‚ β”‚  Desc.  β”‚ β”‚ Results β”‚        β”‚
β”‚  β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜        β”‚
β”‚         β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                β”‚
β”‚                 semua harus FIT + COHERENT                 β”‚
β”‚                                                            β”‚
β”‚  LETHAL TRIFECTA ☠️                                       β”‚
β”‚  β”œβ”€β”€ Data Privat + Komunikasi Eksternal                   β”‚
β”‚  β”‚   + Konten Tak Terpercaya = BAHAYA                     β”‚
β”‚  └── Hapus 1 kondisi β†’ AMAN                              β”‚
β”‚      (contoh: DRAFT email, bukan KIRIM)                   β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

4. Arsitektur Multi-Agent (Capstone) ​

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚  AMPLIFY YOUR BUSINESS - ARSITEKTUR FINAL               β”‚
β”‚                                                          β”‚
β”‚  AGENT 1: BDM (n8n AI Agent)                            β”‚
β”‚  β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”                β”‚
β”‚  β”‚Prospectingβ”‚ β”‚  RevOps  β”‚ β”‚   SDR    β”‚                β”‚
β”‚  β”‚Firecrawl  β”‚ β”‚Pipedrive β”‚ β”‚Gmail     β”‚                β”‚
β”‚  β”‚Hunter.io  β”‚ β”‚Org+Personβ”‚ β”‚Draft     β”‚                β”‚
β”‚  β”‚           β”‚ β”‚+Lead     β”‚ β”‚          β”‚                β”‚
β”‚  β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                β”‚
β”‚         Sub-workflow    Sub-workflow   Sub-workflow      β”‚
β”‚                                                          β”‚
β”‚  AGENT 2: AE (ElevenLabs Voice)                         β”‚
β”‚  β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”                      β”‚
β”‚  β”‚Deal Recordingβ”‚ β”‚Demo Booking  β”‚                      β”‚
β”‚  β”‚Pipedrive     β”‚ β”‚Google        β”‚                      β”‚
β”‚  β”‚Get+Create    β”‚ β”‚Calendar      β”‚                      β”‚
β”‚  β”‚Deal          β”‚ β”‚Avail+Create  β”‚                      β”‚
β”‚  β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜                      β”‚
β”‚       Webhook           Webhook                          β”‚
β”‚                                                          β”‚
β”‚  πŸ”„ LOOP: Outbound emails β†’ Phone callback β†’ Deal+Demo β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

πŸ“Š Seluruh Integrasi Week 3 ​

PlatformCara KoneksiDigunakan Untuk
DockerCLI / DesktopHosting n8n + Ollama
OllamaHTTP (host.docker.internal)AI lokal gratis
OpenRouterAPI KeyAkses DeepSeek, model gratis
Google DriveOAuth2Trigger file drop + download
Google CalendarOAuth2Check availability + create event
GmailOAuth2Create draft outreach emails
FirecrawlAPI + MCPWeb scraping
Hunter.ioMCP (Bearer)Email finder/verifier
PipedriveAPI (Bearer Token)CRM: Leads, Deals, People, Orgs
ElevenLabsVoice Agent + PhoneInbound voice calls
Claude AICustom MCP ConnectorTest n8n MCP Server
PushoverAPIPush notifications

🧠 Pelajaran Terpenting Week 3 ​

#PelajaranKonteks
1Structured Outputs > Tools untuk produksiBoring tapi bulletproof β€” prioritaskan reliability
2Context Engineering = inti AI engineeringBukan tentang model terbesar, tapi input paling tepat
3Lethal Trifecta wajib dicekSetiap workflow: cek 3 kondisi, hapus minimal 1
4MCP = standarisasi koneksi AISatu standar menghubungkan ribuan tools ke AI apps
5Sub-agents = trade-off otonomi vs reliabilityLebih sub-agents = lebih testable, kurang flexible
6LLM = plausible, bukan accurateTugas kita: kontrol, test, evaluasi β†’ jadikan accurate
7Tidak ada resep bakuEksperimen, ukur metrik, pilih yang terbaik β€” ini R&D
8Webhook vs Sub-workflowSub-workflow = internal n8n; Webhook = bisa dari luar (ElevenLabs)
9n8n + ElevenLabs = powerful comboPlatform-platform ini dirancang saling melengkapi
10Build, test, iterateMulai dari business problem + metrik, bukan dari arsitektur

πŸ“ˆ Progres Keseluruhan Kursus ​

Week 1: AUTOMATE   β–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆ 100% βœ…
Week 2: ACCELERATE β–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆ 100% βœ…
Week 3: AMPLIFY    β–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆ 100% βœ…

╔═══════════════════════════════════════════════╗
β•‘                                               β•‘
β•‘   πŸ† KURSUS 100% SELESAI! πŸ†                 β•‘
β•‘                                               β•‘
β•‘   86+ Video Β· 3 Minggu Β· 5 Capstone Projects  β•‘
β•‘                                               β•‘
β•‘   πŸ₯‹ AI Builder BLACK BELT Achieved! πŸ₯‹      β•‘
β•‘                                               β•‘
β•šβ•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•β•

"Remarkably easy to make substantial business impact in a very short amount of time, thanks to the power of n8n and ElevenLabs. Go out there and BUILD!" β€” Instruktur Kursus

Rangkuman kursus Udemy β€” AI Builder: Create Agents, Voice Agents & Automations in n8n