Today we’re launching MCPfy.io — a platform that generates production-ready MCP servers from any repository. Connect GitHub, choose your AI client, and you have a working MCP integration in minutes.
The Problem We Kept Running Into
Every AI-first team we talked to faced the same friction: the gap between “we have an API” and “our AI assistant can use it.” Bridging that gap meant manually implementing the Model Context Protocol — a well-designed but non-trivial open standard that requires schema validation, transport handling, capability negotiation, error codes, and a fair amount of careful testing.
For one service, that’s a reasonable afternoon project. For a team maintaining ten internal services, it becomes a real maintenance burden. And every time an API changes, someone has to update the MCP layer too.
We built MCPfy to automate this translation layer entirely.
How It Works
MCPfy is a three-step pipeline:
1. Static analysis. When you connect a repository, our scanner reads function signatures, route definitions, OpenAPI specs, JSDoc comments, and type annotations across your codebase. It identifies the best candidates for MCP tools, resources, and prompts — the parts of your code an AI can meaningfully interact with.
2. Generation. The generator produces a complete, typed MCP server from the analysis output. Every tool gets proper Zod schemas for input validation. Resources are wired up to the right data sources. The server handles the full MCP lifecycle: initialization, capability exchange, invocation, and shutdown.
3. Deployment. Export directly for local use with Claude Desktop or Cursor. Package as a Docker container. Deploy to Railway, Render, Fly.io, or any cloud provider. The generated config snippets for each AI client are included so installation is a copy-paste.
What We’re Launching
Free tier: Connect one repository, generate up to one MCP server, deploy locally. No credit card, no time limit.
Starter ($19/mo): Five repositories, ten repos, Docker plus one cloud target, email support.
Pro ($49/mo): Unlimited servers and repositories, all deployment targets, ten team members, priority support.
Enterprise: Dedicated infrastructure, custom deployment targets, SLA, and a customer success manager. Contact us for pricing.
Supported AI Clients
The generated servers work with any MCP-compatible client. We’ve tested and include ready-to-use config for:
- Claude Desktop (Anthropic)
- Cursor
- VS Code (via the MCP extension)
- OpenAI (Agents SDK)
- Windsurf (Codeium)
More clients are coming. If your preferred client isn’t listed, the generated server is spec-compliant and should work with any MCP implementation.
The Roadmap
We’re moving fast. Here’s what’s coming in the next few months:
Protocol testing sandbox. Run your generated server in a hosted sandbox environment and validate its behavior against real MCP clients before you deploy. Catches schema mismatches and edge cases before your team hits them.
Incremental sync. When your API changes, MCPfy detects the diff and proposes minimal updates to the generated server rather than regenerating from scratch.
Team workspaces. Shared repositories, role-based access, and audit logs for teams building multiple MCP servers.
More repository hosts. GitLab and Bitbucket are next after GitHub.
More languages. We currently support TypeScript/JavaScript and Python. Go, Rust, and Java are in development.
Building in Public
We’re publishing everything we learn about MCP, AI tooling, and developer platforms on this blog. If you’re building in the MCP ecosystem, we want to hear from you — the best features come from real problems real teams are hitting.
Give MCPfy a try and tell us what you think.