MCP Integration
The Model Context Protocol (MCP) connects your IDE to your DYPAI backend. Your AI agent gets 40+ tools to create tables, endpoints, auth, storage, and more — all from natural language.
Supported AI clients
Works with any MCP-compatible tool. These are the clients we provide ready-made config for:
| Client | Config file | Type |
|---|---|---|
| Claude Code | .mcp.json in project root | Per-project |
| Codex | .codex/config.toml | Per-project |
| Cursor | .cursor/mcp.json | Per-project |
| Windsurf | Global settings | Global |
| Antigravity | Global settings | Global |
| Any MCP-compatible tool | (its own MCP config) | — |
Connect in 2 minutes
Open Conectar IA (MCP tokens)
From your organization, open the Conectar IA (MCP tokens) page at /{org}/mcp/tokens and create a token.
Copy and paste
Copy the JSON config and paste it into your client's MCP config file. The dashboard always shows the exact server URL — copy it from there rather than typing it by hand:
{
"mcpServers": {
"dypai": {
"type": "http",
"url": "https://mcp.dypai.dev/mcp",
"headers": {
"Authorization": "Bearer dypai_mcp_YOUR_TOKEN"
}
}
}
}
Verify
Ask your AI: "What tables are in my project?" — if it responds with your tables, you're connected.
Install with CLI
You can configure MCP without copying JSON manually. On a machine with Node.js installed, run:
npx @dypai-ai/install
The installer lets you pick your client (Claude Code, Codex, Cursor, Windsurf, Antigravity, and others) and can open a browser to sign in and fetch a token if you do not pass one. Non-interactive example:
npx @dypai-ai/install --client cursor --token dypai_mcp_YOUR_TOKEN
Run npx @dypai-ai/install --help for all options.
Starter templates
Download a template from the Frontend tab — it comes with the MCP config pre-configured for your IDE.
Remote vs local MCP
DYPAI ships MCP in two flavors. Most people use both:
| Remote (cloud) MCP | Local MCP — @dypai-ai/mcp | |
|---|---|---|
| What it is | A hosted HTTP server you connect to | An npm package that runs on your machine |
| Tools | 41 agent-facing tools | 28 tools (default profile) |
| Best for | In-IDE editing — tables, endpoints, auth, logs, etc. | Filesystem-aware deploy — push/pull your code, deploy to production |
| Setup | Paste the JSON config above | npx @dypai-ai/install |
The JSON config above connects you to the remote server for editing. The local package adds filesystem-aware deploy on top — it reads your project files to ship them. Read the local package guide.
What can the AI do?
Once connected, just describe what you need:
- "Create a users table with email, name, and role"
- "Build a POST endpoint that saves orders and sends an email"
- "Set up Google OAuth"
- "Add a Stripe payment node to the checkout workflow"
- "Show me the version history of the orders endpoint"
- "Rollback the checkout endpoint to yesterday's version"
Security
| Feature | Detail |
|---|---|
| Scoped tokens | Each token is scoped to one project. The AI only sees that project |
| Audit log | Every MCP action is logged with timestamp and details |
| Revocable | Regenerate or revoke tokens anytime from the dashboard |
| Encrypted | Tokens are encrypted at rest in the database |