AgentForge is a unified API gateway and subscription marketplace for AI agents, positioned as "one API key, access to 300+ agents, zero configuration." It is designed to solve the problem that most agent platforms require separate management of API keys, authentication flows, and billing for each agent.
Purpose and positioning: To provide developers and AI agents with a unified agent invocation and distribution platform, which can be used to invoke a large number of agents through a single entry point, and can also be integrated by MCP-compatible clients as an MCP server.
Main capabilities:
- Unified API: Invoke any agent through a single REST endpoint.
- 300+ agents: Preloaded popular agents from GitHub and HuggingFace.
- Creator economy: Publish your own agents and set prices; creators receive a 90% revenue share (the platform takes 10%), with withdrawals to bank accounts via Stripe Connect.
- For developers: Provides RESTful API, streaming support, API key authentication, and rate limiting.
- MCP support: Built-in MCP server (mcp/server.ts), exposing all 300+ agents as MCP tools, so that MCP clients supporting stdio transport, such as Claude Desktop, Cursor, and Continue, can discover and invoke them with zero additional configuration.
- Platform capabilities: Google OAuth and email/password authentication, TOTP two-factor authentication, rate limiting (1,000 requests per key per hour, 10,000 per day), agent health monitoring, and automatic import from GitHub trends and HuggingFace.
- Usage and billing: Supports usage tracking and billing history.
How to use:
- Use the API directly (no installation required).
- Use via the MCP server: Add the configuration in Claude Desktop's claude_desktop_config.json (on macOS the path is ~/Library/Application Support/Claude/claude_desktop_config.json) and restart to see the AgentForge tools; other MCP clients supporting stdio can also connect.
- Self-hosting: Requires Node.js 20+ and PostgreSQL. The steps are: git clone https://github.com/doggychip/agentforge.git、cd agentforge, npm install, set the environment variable DATABASE_URL, then npm run dev (automatic migration and data seeding).
- Deploy to Zeabur: Push to GitHub, create a project in Zeabur, import the repository and add a PostgreSQL service; Zeabur automatically injects DATABASE_URL.
API endpoints:
- POST /api/agents/:id/invoke: Invoke an agent
- GET /api/agents: List all agents
- GET /api/agents/:id: Get agent details
- GET /api/agents/:id/health: Check agent health status
- GET /api/stats: Platform statistics Full API documentation: patreon.zeabur.app/#/docs
Environment variables: DATABASE_URL (required, PostgreSQL connection string); STRIPE_SECRET_KEY, STRIPE_WEBHOOK_SECRET, GOOGLE_CLIENT_ID, GOOGLE_CLIENT_SECRET, SMTP_HOST, SMTP_USER, SMTP_PASS (all optional, used for payments, OAuth, and email).
Tech stack: The frontend uses React 18, Tailwind CSS, shadcn/ui, TanStack Query, and wouter; the backend uses Express 5, Drizzle ORM, and Passport; the database is PostgreSQL; payments use Stripe Connect; authentication uses bcrypt, Google OAuth, and TOTP 2FA; deployment uses Docker / Zeabur; MCP uses @modelcontextprotocol/sdk (TypeScript).
Project structure: client/src/ is the React frontend (pages, components, hooks); mcp/server.ts is the MCP server (5 tools over stdio); server/ contains routes.ts, storage.ts, and db.ts; shared/schema.ts contains the Drizzle schema and types; it also includes a Dockerfile.
Limitations and notes: The project uses the MIT license, pull requests are welcome, and for major changes it is recommended to open an issue first. The specific number of agents, pricing, revenue share ratio, etc. are subject to official materials; information not provided in the official materials is not yet available.
