Powering Compose Multiplatform Apps with Koog & MCP
summary
Live coding builds a Compose Multiplatform app that runs a Koog agent wired to real data: a Ktor client + local Kotlin tools call ClimateTrace, then a Kotlin MCP SDK server exposes the same tool over STDIO and Cloud Desktop. Demos run on JVM, Android and iOS; local LLMs include a 35B Mixture‑of‑Experts model on desktop and a 2B Gemma on Android via LightORT. A custom conversational strategy uses a channel to unblock on-assistant replies, telemetry lands in LangFuse, and the flow largely works with occasional LLM nondeterminism. Kotlin MCP SDK Ktor client
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:05LLMs and AI agents
- 3:11What is Kook and KMP
- 5:16Live coding session
- 7:49Ktor client for real data
- 10:50Local tools for agent
- 13:00Compose Multiplatform UI
- 19:06Custom conversational strategy
- 28:05Telemetry with LangFuse
- 30:16Local models on JVM/Android
- 36:12MCP server with Kotlin SDK
- 39:02Testing MCP server
- 39:45Connecting to Cloud Desktop
- 43:01Terminal UI demo
- 44:28Summary and Q&A
related talks
48mAndroid, Compose, Multiplatform & Server - Inject them all with Koin
40mBuilding a Modern Kotlin Multiplatform Application with Compose Multiplatform and Koin
45mThe Adventures of Kotlin And Compose Through the Multiplatform World
32mHitchhiker's Guide to Kotlin/Compose Multiplatform samples and libraries
44mThe Adventures of Kotlin & Compose Through the Multiplatform World
38mDesign and Implementation of Next‑Gen App Development Tools Using Compose Multiplatform × AI
42mWhat’s New in Compose Multiplatform
41mKotlin for the JVM and beyond