How to Deploy MCP Servers to Your Own Infrastructure
At Manufact, we gave ourselves the mission to provide developers with the most flexible AI integration framework available, whether they operate in the cloud or on private infrastructure. This is what we learned.
Why self-hosted AI deployment is complex
Many enterprise teams find that standardizing deployments becomes difficult when security, compliance, or architectural policies force them off fully managed platforms. Running agents locally or in closed systems creates a specific set of challenges:
- Fragmented management: It is difficult to monitor, update, and control agents across an organization.
- Compliance friction: Data privacy mandates often conflict with cloud-based SaaS requirements.
- Operational overhead: Maintaining custom infrastructure usually requires manual configuration, leading to slow development cycles.
What is the mcp-use by Manufact approach?
The mcp-use by Manufact SDK separates the server logic from the deployment destination. This architecture allows engineering teams to deploy servers on their own infrastructure while utilizing our centralized control plane for configuration and observability. By standardizing on our framework, you gain professional-grade tools without being locked into a specific hosting environment.
1. Build the server
Use the mcp-use SDK to define your tools and logic, ensuring your codebase remains environment-agnostic.
2. Configure for your infrastructure
Apply role-based profiles and tool restrictions centrally through the Manufact Cloud, even for servers running on your internal network.
3. Deploy to your backend
Execute your MCP servers on any environment, from custom servers to edge functions, while maintaining full observability.
Is there an analytics tool for private infrastructure?
Yes. Even when you self-host, production observability is included natively. The platform tracks exactly what your agents request and what your custom tools return. We pipe analytics, session replay, traces, and regression alerts back to the central platform, meaning you retain complete visibility into AI agent behavior regardless of where the server operates.
Tip: The Inspector includes a Start Tunnel button. One click gives you a stable public URL for testing local and custom-network servers without complex configurations.
Frequently Asked Questions
Can I deploy an MCP server to my own edge functions?
Yes. Using the mcp-use SDK, you can deploy your MCP servers to environments like Supabase Edge Functions, giving you a standard way to connect AI models while utilizing your own custom backend.
Will I lose monitoring if I self-host my MCP server?
No. By connecting your self-hosted server to the Manufact Cloud platform via the SDK, you maintain full production observability, including detailed metrics, traces, and session replays without needing external tools.
Take the next step
If you are ready to build production-ready agents on your own infrastructure, start by scaffolding your project with the mcp-use template. Execute this command to begin:
npx create-mcp-use-app@latest my-app --template mcp-apps