How I Built a Live Streaming Platform with OSC + MCP
A hands-on tutorial showing how AI agents can discover and compose OSC services through MCP to build complete solutions. From zero to production in a single conversation.
Model Context Protocol (MCP) enables AI agents to interact with external systems in standardized ways. When combined with Open Source Cloud's discoverable service catalog, this creates a development experience unlike anything else: describe what you want to build, and watch AI discover, provision, and orchestrate the infrastructure to make it happen. This tutorial walks through building a complete live streaming platform using nothing but natural language conversation.
What We'll Build
Our live streaming platform will accept RTMP input streams, transcode them to multiple bitrates, package them for adaptive bitrate delivery using HLS, store recordings in object storage, and serve streams through a CDN. This would traditionally require weeks of infrastructure work, service integration, and configuration debugging. With OSC and MCP, we will build it through conversation in under an hour.
Setting Up MCP with Claude
First, install Claude Code CLI from Anthropic. Then configure the OSC MCP server by adding it to your Claude configuration file at ~/.config/claude/config.json. You will need your OSC Personal Access Token, which you can generate from the OSC dashboard under Account Settings. Once configured, Claude can now query the OSC service catalog, read API documentation, provision services, and manage your infrastructure through natural language. No API documentation reading required — Claude discovers capabilities automatically.
Building Through Conversation
Start a conversation with Claude Code: "I want to build a live streaming platform that accepts RTMP input, transcodes to multiple bitrates, packages as HLS, and stores recordings in object storage. What OSC services should I use?" Claude will query the OSC catalog, analyze available services, and recommend a stack: SRS (Simple Realtime Server) for RTMP ingest, SVT Encore for transcoding, Shaka Packager for HLS packaging, MinIO for object storage, and NGINX for serving streams. This recommendation is not hardcoded — Claude discovered these services by querying the catalog and reading their documentation.
Automated Service Discovery
What makes this powerful is that Claude is not working from a fixed list of services. When you ask Claude to provision the recommended services, it uses MCP to query each service's API schema, read configuration documentation, provision instances with appropriate settings, and configure service interconnections automatically. There is no manual configuration file editing, no hunting through documentation, and no trial-and-error debugging. The AI handles the orchestration by discovering the correct APIs and configuration patterns in real-time.
From Conversation to Deployment
Once services are provisioned, Claude can generate the application code that orchestrates them. Ask Claude to create a web interface for stream management, and it will build a Next.js application with forms for creating streams, real-time status monitoring, and playback controls. Ask Claude to add authentication, and it will integrate OAuth 2.1 with your OSC account. Ask for monitoring and analytics, and it will add appropriate logging and metrics collection. Each request triggers service discovery, code generation, and deployment — all through conversation. When you are done, you have a complete, production-ready live streaming platform running on OSC infrastructure.
What Makes This Unique
No other cloud platform offers this workflow. Traditional platforms require you to learn their specific services, configuration formats, and deployment processes. Infrastructure-as-code tools require you to write Terraform or CloudFormation templates. Even AI-assisted coding tools can only generate code — they cannot provision infrastructure. OSC with MCP is the only platform where AI can discover services, understand their capabilities, provision infrastructure, and generate orchestration code through natural language. This is not vibe coding where you describe code and get code back. This is conversational orchestration where describing requirements produces running systems.
Try It Yourself
Ready to build through conversation? Install Claude Code, configure the OSC MCP server, and start describing what you want to build. The full setup instructions are available in our documentation.