Skip to main content
Magic Resume is an open-source, AI-native resume platform built around three ideas:
  • Local-first by default — all resume data lives in IndexedDB. No account or backend is required to run the app. Cloud sync is opt-in.
  • Schema-first templates — every resume is a typed Zod value. Templates render from a shared schema, and AI tools edit resumes via JSON Patch against that schema (no guessing).
  • AI tools as first-class citizens — a native MCP server (@magic-resume/mcp) lets Claude Code, Cursor and Windsurf read and safely edit your resumes through the same data layer the web app uses.
This site is the contributor and integrator handbook. If you want to use the hosted app, see magic-resume.cn. If you want to run it, hack on it, or wire AI tools into it, you are in the right place.

​
System architecture

This repo holds the frontend, the MCP server, and the shared schema. The backend runs as a separate deployment. The frontend talks to one gateway origin, which routes /api/* by path prefix to the backend services. Magic Resume system topology

​
1. Magic-Resume (this repo)

github.com/LinMoQC/Magic-Resume — pnpm + Turborepo monorepo: Responsibility: all user-facing UI, local editing, IndexedDB persistence, MCP tooling. The web app also carries its own Next.js route handlers under app/api/* (e.g. chat-agent, pdf/parse) that authenticate the user and proxy AI requests to the backend.

​
2. The backend (separate deployment)

A NestJS pnpm-workspace monorepo, split by runtime profile and fronted by one gateway:
  • gateway — the single public entry point. Edge auth + trusted identity injection, per-IP/user rate limiting, and streaming reverse-proxy routing of /api/* by prefix. Zero DB.
  • platform-api — CRUD: resumes, users, sharing/collaboration, notifications, JSON Patch application, Zod validation. Owns PostgreSQL.
  • agent-service — the AI backend: DeepAgents-driven chat / workflows / interview (LangGraph), streamed over SSE; PDF parse; translation.
Why it’s separate: the browser-side React tree and the server-side database are different concerns. Splitting the backend out means the MCP server (and future third-party clients — mobile, CLI) share one backend without dragging the frontend along.

​
Contracts

Schema consistency is enforced through @magic-resume/resume-schema — published to npm and consumed by the backend as a dependency. Defined once, validated everywhere.

​
Deployment modes

  • Self-hosted: only apps/web needs to run. Data stays in IndexedDB — resume editing and export only. No account, no backend, no AI (AI needs the backend / cloud mode). See Self-hosted.
  • Cloud (magic-resume.cn): frontend and backend deploy separately — Web on Vercel, the backend on a container platform. The frontend points NEXT_PUBLIC_API_URL at the gateway. See Cloud mode.

​
Tech stack

​
Where to go next

Getting Started

Architecture overview

MCP server

Local development

This documentation tracks the master branch. If you are working on a feature branch and something here is out of date, the CLAUDE.md at the repo root is the canonical short-form reference.