Hey PH! I'm Tejas, and I built Platos because I kept hitting the same wall with every AI agent project: the agent runtime was 10% of the work, and the other 90% was auth, memory, tool registry, monitoring, rate limits, and deployment plumbing. Every framework gives you the agent loop. None of them give you the ops layer around it. FastMCP is Flask. Platos is Heroku. That's the line we drew. The whole thing is open source (MIT), ships via Docker Compose, and runs on your own infra with your own API keys (BYOK — we never touch your model billing). The Python and TypeScript SDKs let you turn any function into an MCP tool with one decorator, and the platform handles everything else: - Three-tier memory with sub-20ms reads - Smart tool selection with no LLM in the path - MCP gateway with OAuth 2.1 - Multi-agent orchestration with a visual builder - Full Next.js dashboard for the whole thing You can try the live demo right now at test.platos.dev (login: demo@platos.dev / Demo@Platos) or clone the repo and run make up. Would love your feedback!

Platos

Turn any function into a production MCP tool in 60 seconds

Turn any function into a production MCP tool in 60 seconds