RepoFuse MCP Server

RepoFuse Backend

An AI-powered code intelligence platform that scans your GitHub repositories and discovers what applications you can build by combining existing files and components.

Features

• GitHub App Auth: Connect your GitHub account with a single click using GitHub App user authorization
• Repository Management: Add and manage GitHub repositories for analysis
• AI Code Analysis: AI scans every file to identify purpose, exports, and reusability
• App Blueprint Discovery: Discover applications you can build from your existing code
• Gap Analysis: See exactly which files you're missing and generate them with AI
• Export: Download blueprint JSON for offline use or share with your team
• Stripe Billing: Real checkout flow for RepoFuse Pro upgrades and billing management

Growth Strategy

• Developer SaaS Growth Brief: developer-led distribution, high-intent GitHub outreach, Product Hunt timing, first-user tests, and RepoFuse's first 14-day growth plan.

Tech Stack

• Framework: Next.js 16 with App Router
• Database: Neon PostgreSQL with connection pooling
• AI: Vercel AI SDK (OpenAI GPT-4)
• UI Components: Shadcn UI with Radix primitives
• Styling: Tailwind CSS v4
• Auth: GitHub App user authorization (custom)

Project Structure

app/
├── api/                              # API Routes
│   ├── auth/github/callback/         # GitHub App callback
│   ├── github/repos/                 # Fetch user's GitHub repos
│   ├── github/create-repo/           # Create repo from blueprint
│   ├── repositories/                 # Repository CRUD
│   ├── analyses/                     # Analysis management
│   │   └── [id]/
│   │       ├── run/                  # Run analysis (SSE stream)
│   │       └── analyze/              # AI analysis endpoint
│   └── export/
│       ├── blueprint/                # Export blueprint as JSON
│       └── pdf/                      # Export blueprint as PDF
├── dashboard/                        # Dashboard pages
│   ├── layout.tsx                    # Dashboard layout
│   ├── page.tsx                      # Overview
│   ├── repositories/                 # Repository management
│   └── analyses/                     # Analysis pages
│       └── [id]/                     # Analysis detail
components/
├── repositories-list.tsx             # Repository list + add form
├── repository-selector.tsx           # Multi-repo selector
├── analyses-list.tsx                 # Analyses list
├── analysis-detail.tsx               # Analysis results + blueprints
├── app-suggestions.tsx               # App idea cards
└── ui/                               # Shadcn components
lib/
├── db.ts                             # Neon database client
├── queries.ts                        # Database queries
└── utils.ts                          # Utility functions
scripts/
└── 01-create-schema.sql              # Database migration

Database Schema

user_auth

• github_id: Unique GitHub user ID
• github_username: GitHub login name
• github_avatar_url: Profile picture URL
• access_token: GitHub user access token
• stripe_customer_id: Stripe customer linked to the user
• stripe_subscription_id: Active or latest Stripe subscription id
• stripe_price_id: Stripe price id for the user’s current plan
• plan_tier: free / pro
• subscription_status: Stripe subscription status

repositories

• github_id: Unique GitHub repo ID
• name, full_name: Repo name and owner/name
• description, url: Metadata
• default_branch, language, stars: Additional info

repo_files

• repository_id: Foreign key to repositories
• path, name, extension: File location info
• file_type: component / hook / utility / api / page / etc.
• purpose, ai_summary: AI-generated descriptions
• technologies, exports, imports: Detected patterns (JSONB)
• reusability_score: 0–100 reusability rating

analyses

• name: User-defined analysis name
• status: pending / scanning / analyzing / complete / failed
• total_files, analyzed_files: Progress tracking

analysis_repositories

• Junction table linking analyses to repositories

app_blueprints

• analysis_id: Foreign key to analyses
• name, description, app_type: Blueprint info
• complexity: simple / moderate / complex
• reuse_percentage: How much existing code can be reused
• existing_files, missing_files: File lists (JSONB)
• estimated_effort: Human-readable time estimate
• technologies: Detected stack (JSONB)
• ai_explanation: AI-written rationale

Getting Started

1. Clone the repository

git clone <repository-url>
cd repofuse-backend

1. Install dependencies

pnpm install

1. Set up environment variables Copy .env.example to .env.local and fill in your values:

cp .env.example .env.local

1. Set up the database Run the schema migration in your Neon console or with psql:

psql $DATABASE_URL -f scripts/01-create-schema.sql

1. Run the development server

pnpm dev

1. Access the application Open http://localhost:3000 in your browser

RepoFuse MCP Server

This repo now includes a standalone stdio MCP server at mcp/repofuse.mjs.

What it exposes

• list_github_repositories
• analyze_repositories
• generate_scaffold
• create_repo_from_blueprint
• hosted only: get_blueprint_gaps

Environment variables

The MCP server expects:

• GITHUB_TOKEN
• AI_GATEWAY_API_KEY or VERCEL_OIDC_TOKEN, otherwise ANTHROPIC_API_KEY
• optional: REPOFUSE_MODEL (anthropic/claude-sonnet-4.6), REPOFUSE_MAX_FILES_PER_REPO, REPOFUSE_MAX_BLUEPRINTS

Run it locally

pnpm mcp:repofuse

Smoke-test the MCP server

pnpm mcp:test

Use the live variant when you want to verify startup with real credentials:

pnpm mcp:test:live

Streamable HTTP endpoint inside RepoFuse

RepoFuse exposes an OAuth-protected Streamable HTTP endpoint at https://repofuse.com/api/mcp. It supports external assistants and authenticated web-app sessions, resolves GitHub access server-side, and enforces account plan, credit, and rate limits.

Connection instructions: https://repofuse.com/mcp.

Example Claude Desktop config

See examples/claude-desktop.mcp.json.

Example Cursor config

See examples/cursor.mcp.json.

Cursor plugin config

See .cursor-plugin/plugin.json and the root mcp.json.

Full MCP setup guide

See docs/MCP_SETUP.md and docs/LISTING.md.

API Endpoints

Authentication

• GET /api/auth/github/callback - GitHub App callback

Repositories

• GET /api/repositories - List tracked repositories
• POST /api/repositories - Add repository by URL
• GET /api/repositories/[id] - Get repository details
• DELETE /api/repositories/[id] - Remove repository

GitHub

• GET /api/github/repos - Fetch repos available to the signed-in GitHub App user
• POST /api/github/create-repo - Create new repo from blueprint (Pro)

Billing

• GET /pricing - Pricing page
• GET /api/checkout?plan=pro - Start Stripe checkout for Pro
• GET /api/checkout/success - Finalize successful checkout and sync subscription
• GET /api/billing/portal - Open Stripe billing portal

Analyses

• GET /api/analyses - List analyses
• POST /api/analyses - Create new analysis
• POST /api/analyses/[id]/run - Run analysis (Server-Sent Events)
• POST /api/analyses/[id]/analyze - AI pattern analysis

Export

• POST /api/export/blueprint - Export blueprint as JSON
• POST /api/export/pdf - Export blueprint as PDF

Deployment

Deploy to Vercel

1. Push your code to GitHub
2. Connect your repository to Vercel
3. Add environment variables in Vercel dashboard (see .env.example)
   • For paid upgrades, create a recurring Stripe Price for RepoFuse Pro and set STRIPE_SECRET_KEY and STRIPE_PRO_PRICE_ID
4. Deploy

Security

• GitHub App permissions are fine-grained and should be configured read-only for analysis access
• Access tokens are stored in the database (encrypt at rest in production)
• Code is scanned in memory; file contents are never permanently stored
• Hosted MCP requests validate Clerk OAuth tokens or an existing RepoFuse session
• OAuth discovery follows RFC 9728 and GitHub tokens are never returned to MCP clients

License

MIT License