Let Claude build your backend (MCP)

APIForge ships a native MCP server (Model Context Protocol), the open standard AI assistants use to operate tools. Connected, your assistant gets 26 tools: create projects, model resources, wire relationships, enable auth, configure localization, import data, mint API keys.

1 · Get your personal MCP URL

Open My MCP in the panel. Every user gets a unique URL: everything your AI creates is owned by your account and scoped to your projects. (An admin enables MCP per user; it's off by default.)

2 · Connect your client

ChatGPT: Settings → Connectors → Add custom connector → paste the URL.
Cursor: Settings → MCP → Add server (HTTP URL).
Claude Desktop: add an mcp-remote entry with your URL; the snippet is pre-generated on the My MCP page.

3 · Describe what you need

“Create a project for a recipe site. Resources: recipes (title, ingredients, steps, photo, category) and categories. Link them. Enable login with email OTP. Add Swedish alongside English. Seed 5 sample recipes and give me an API key and the docs URL.”

The assistant calls create_project → create_endpoint → create_relationship → setup_auth → configure_project → seed_resource_records → create_api_key and replies with your live URLs.

Why it doesn't hallucinate your stack

The first thing a well-behaved assistant calls is describe_platform, a machine-readable reference of every field type, relationship kind, auth option and settings group. The AI learns the real vocabulary before building, and every write is validated server-side against your schema anyway.

Iterate in plain language

“Add a prep_time number field to recipes.” “Make categories publicly readable.” “Turn on draft preview.” Each is one tool call, visible in your panel immediately, with the same permissions and validation as the UI.