MCPify is the MCP server generator that takes your OpenAPI spec, database schema, or whole codebase and produces a working MCP server in seconds — so your AI agents can finally call your API without you writing a line of MCP boilerplate.
Let's be honest: writing an MCP server by hand is a slog. You sit there mapping every endpoint to a tool definition, hand-rolling JSON schemas, describing parameters so Claude or GPT actually knows what to do with them, and then re-doing half of it when your API changes. It's slow, it's error-prone, and it's the exact kind of busywork that should've been automated years ago.
And it gets worse the moment your API grows. Every new endpoint means a new tool. Every schema change means drift between what your API actually accepts and what your agent thinks it accepts. Before you know it, your agent is hallucinating parameters that don't exist, or failing on endpoints you forgot to expose. That's not a "fun MCP project" anymore — that's a maintenance tax you pay every single sprint.
Here's the thing though: you already have everything needed to describe your API to an agent. Your OpenAPI spec. Your Prisma schema. Your routes. You just need an MCP server generator to do the boring part for you. That's exactly what MCPify is.
MCPify is an open-source AI enablement compiler — think of it as the MCP server generator for the rest of us. Built by Amarnath and maintained as a community project on GitHub, it scans the parts of your application that matter to agents and compiles them into a runnable MCP server, complete with typed tools, permission-aware boundaries, and descriptions that agents actually understand.
It's a single npm package (mcpify-cli) with no installation required — you run it with npx, point it at your project, and it does the analysis and generation for you. The philosophy is simple: your app stays untouched; MCPify just reads it and emits the MCP layer.
You have a spec. MCPify has a command. That's the whole transaction. Running the swagger command converts your OpenAPI or Swagger spec directly into MCP tools:
npx mcpify-cli swagger ./tests/fixtures/swagger/petstore.yamlInstead of hand-mapping each endpoint to a tool definition, you get a complete MCP surface with typed inputs, descriptions, and metadata — ready to register with Claude Code, Claude Desktop, Codex, or VS Code via the --clients flag.
MCPify doesn't just read a spec; it reads your whole stack. The default analyze command runs backend analysis plus optional OpenAPI, Prisma, Drizzle, and Mongoose analysis in one pass. For an ecommerce stack with Prisma and a Swagger spec:
npx mcpify-cli analyze ./examples/ecommerce-saas \
--output ./examples/ecommerce-saas/.mcpify \
--prisma ./examples/ecommerce-saas/prisma/schema.prisma \
--swagger ./examples/ecommerce-saas/openapi.jsonFrontend got a React or Vue UI? The frontend analyzer maps UI actions to agent-controllable operations too. Workflows like checkout get exposed as atomic, composable steps rather than a pile of raw endpoints.
Every generated tool passes through a permission layer (scopes, roles, audit trails) before it touches your system, and with --ai-enhance MCPify uses an LLM to improve tool descriptions so agents stop guessing. There's even an audit command for a static safety review and a simulate command to run AI simulations against your compiled tool surface before you ship it.
I get asked this a lot, so let's be direct about it.
| MCPify | Hand-rolled MCP | Inngest / Temporal | |
|---|---|---|---|
| Generates an MCP server from your API | Yes, automatically | You write every tool | Not their focus |
| OpenAPI / Prisma input support | Built in | Manual | — |
| Keeps tools in sync with code | Regenerates on commit | Manual updates | — |
| Core job | Expose your app to agents | Agent tooling | Workflow orchestration & durability |
To be fair to Inngest and Temporal: they're genuinely great at what they do — durable workflows, retries, and orchestration at scale. But they're not MCP server generators. They don't look at your OpenAPI spec and emit an MCP server, and they don't solve the tool-drift problem. If you're building agent workflows, you might well pair one of them with MCPify rather than pick between them. And compared to hand-rolling MCP servers? There's no comparison. Hand-rolling means you own every schema, every description, every sync problem forever. MCPify means you own one command. I know which one I'd rather maintain.
You don't need to install anything. From the root of your app:
npx mcpify-cli analyze ./my-appThat's it. The pipeline runs backend analysis, optional schema and spec analysis, event discovery, workflow detection, permission classification, and MCP server generation. The generated AGENTS.md explains exactly how to connect the compiled server to your MCP client. Want to start from an OpenAPI spec alone?
npx mcpify-cli swagger ./openapi.jsonTwo commands, and your API speaks MCP. That's the whole pitch.
Backend engineers who'd rather ship features than MCP boilerplate. Platform teams that need to expose internal APIs to AI agents without a bespoke integration per service. Founders and indie hackers who want their product to be "AI-ready" without rewriting it. And honestly, anyone who's ever stared at an OpenAPI spec and dreaded writing the 40th tool definition. If you build APIs and want agents to use them, MCPify is the MCP server generator that meets you where you are.
Usually seconds. MCPify's swagger command reads your spec and emits a typed MCP server immediately — no hand-mapping, no JSON-schema archaeology. For a full codebase analysis with schema and workflow detection, expect a couple of minutes on a typical app.
Yes. The backend analyzer is framework-aware for Express, Fastify, NestJS, and Next.js, and it handles both JavaScript and TypeScript codebases — including inline, non-exported handlers that naive scanners miss.
Absolutely. That's the point. If you don't have a spec, MCPify analyzes your actual code — routes, controllers, services, Prisma/Drizzle/Mongoose schemas, even frontend actions — and generates the MCP server from what it finds.
Yes, it's open source and free. The CLI runs locally via npx, and the code lives on GitHub. The only optional paid piece is the AI enhancement flag, which uses your own ANTHROPIC_API_KEY when you pass --ai-enhance.
Any MCP-compatible client. MCPify can auto-register into Codex, Claude Code, Claude Desktop, and VS Code with the --clients flag, or you can connect the server to anything else that speaks MCP.
Hand-writing MCP servers is a solved problem with a better answer. MCPify reads what you already have — your OpenAPI spec, your schemas, your code — and compiles it into an MCP server your agents can actually use, safely and without drift. No boilerplate, no rewrites, no endless maintenance.