Why Host MCP Server on EdgeOne Makers
The platform's edge functions and KV storage provide the ideal architecture for hosting mcp server — Streamable HTTP transport, custom domains, global edge distribution, and OpenAI-compatible API format for all mcp hosts.
One-Command Deployment
Deploy with `git push` for auto-deploy or `edgeone makers deploy` via CLI — no Docker, no Kubernetes, no server provisioning.
Built-in Agent Runtime
Managed runtime with session-sticky routing, up to 1-hour execution time, and in-memory state reuse — purpose-built for LLM calls and multi-step agent loops.
Integrated AI Models
Access DeepSeek, MiniMax, Hunyuan, and more through a unified AI gateway. New accounts receive 500K free tokens with zero configuration.
Full Observability
Zero-instrumentation distributed tracing — view complete call chains, LLM interactions, tool invocations, and latency metrics in local and cloud dashboards.
How to Host MCP Server in 3 Steps
Implement MCP protocol on edge functions, configure KV storage, and bind a custom domain — or start from the MCP On Edge template. See the MCP documentation: https://pages.edgeone.ai/document/mcp for a complete guide.
1
Write Your Agent
Build in the `agents/` directory with any framework (OpenAI SDK, Claude SDK, LangGraph, CrewAI, DeepAgents).
2
Deploy to Production
Push to Git (GitHub/GitLab/Gitee) for auto-deploy, or run `edgeone makers deploy` via CLI.
3
Go Live
Deploys globally in minutes with automatic SSL and edge routing.
Platform Capabilities for MCP Hosting
How EdgeOne Makers enables hosting mcp server — edge functions for MCP protocol, KV storage for data, custom domains for endpoints, and global distribution for low-latency mcp hosts.
Feature | Description |
|---|---|
| Agent Runtime | Hosts LLM calls, Agent loop orchestration, and business logic, with session-based routing and automatic scaling. |
| Sandbox Tools | Provides two separate yet interoperable API layers for both LLMs and developers. Browser automation, code execution, Shell, and file operations all run in an isolated sandbox environment. |
| Conversation Storage | Provides framework-compatible memory management, with unified APIs for sessions and messages. |
| Observability | Automatically collects call traces with zero-intrusion instrumentation, enabling unified trace viewing in both local and cloud dashboards. |
| Built-in Models | Access Hunyuan and other mainstream Chinese models through AI Gateway with a limited-time free token quota. |
What platform capabilities do I need to host mcp server?
Edge functions for MCP protocol implementation (Streamable HTTP transport), KV storage for data persistence, and custom domain binding for client endpoints. EdgeOne Makers provides all three. See the MCP documentation: https://pages.edgeone.ai/document/mcp for details.
Can I build a custom MCP server or must I use the template?
You can build from scratch on edge functions or start from the MCP On Edge template (MIT, open source). The template includes MCP Server + MCP Client + Backend API components — customize freely for your mcp hosts.
Is hosting mcp server free on EdgeOne Makers?
Yes. Edge functions and KV storage are included in the free tier. The MCP On Edge template is open source (MIT). No credit card required to start hosting mcp server.
Which AI clients can connect to my hosted MCP server?
Any client supporting Streamable HTTP MCP Server (2025-03-26 spec): Cursor, Claude Desktop, Cline, VSCode, Windsurf, ChatWise. Configure with your custom domain endpoint URL.
Can my MCP server serve both MCP protocol and OpenAI-compatible API?
Yes. The platform's edge functions support multiple routes — your mcp hosts can expose `/mcp-server` for MCP clients and `/v1/chat/completions` for OpenAI-compatible requests simultaneously.