# Create custom MCP server to Query Supabase - without code

Supabase has become the default backend for modern startups and indie hackers. It gives you a full Postgres database, authentication, and real-time subscriptions out of the box.

But as you build AI features—like an agent that manages user subscriptions or moderates content—you hit a wall. You need to give your AI secure access to your Supabase data. You don't want to hardcode your `service_role` key into an LLM prompt.

**Pylar is the missing link.**

Pylar allows you to create a secure, read-only MCP (Model Context Protocol) server for your Supabase project in minutes. It respects your data privacy and gives you a governance layer for your AI agents.

## Why Pylar for Supabase?

- **Native Postgres Support:** Pylar connects directly to Supabase's underlying Postgres instance.
- **Connection Pooling:** Pylar works with Supabase's transaction pooler (PgBouncer) for high-scale apps.
- **Security:** Keep your `service_role` keys secret. Pylar uses standard database credentials and IP whitelisting.

## Step-by-Step Walkthrough

### Step 1: Get Connection Details from Supabase

1. Log in to your **Supabase Dashboard**.
2. Go to **Project Settings** -\> **Database**.
3. Under **Connection parameters**, uncheck "Use connection string" to see the individual fields.
4. Copy your **Host**, **Database Name**, **User**, and **Port** (usually 5432 or 6543).  
   - _Note:_ You will need your database password. Supabase doesn't show this, so hopefully you saved it!

### Step 2: Connect Supabase to Pylar

1. In Pylar, go to **Connections** -\> **PostgreSQL** (Supabase is just Postgres!).
2. Enter the details you copied: **Host**, **Port**, **Database**, **Username**, and **Password**.  
   - **Naming Rule:** Use lowercase letters and underscores (e.g., `supabase_prod`).

1. **Whitelist IP:** In Supabase, go to **Network Restrictions** (under Database settings). Add Pylar's IP: `34.122.205.142`. Pylar will verify the connection.

### Step 3: Create a Sandboxed View

Let's say you want an agent to help with customer support, but you need to protect user privacy.

In Pylar's SQL IDE, create a view:

```sql
-- view: user_subscription_status
SELECT
  id as user_id,
  email,
  raw_user_meta_data->>'full_name' as name,
  subscription_tier,
  last_sign_in_at
FROM auth.users
JOIN public.subscriptions ON auth.users.id = public.subscriptions.user_id
WHERE status = 'active'
```

This view joins Supabase's internal `auth` schema with your public `subscriptions` table, giving the agent exactly what it needs and nothing more.

### Step 4: Auto-Build the MCP Tool

1. **Select your view** (`user_subscription_status`) in the right sidebar.
2. Click **"Create MCP Tool"** and choose **"Create with AI"**.
3. Type a prompt: _"Create a tool that gets a user's subscription tier by email."_
4. Pylar generates the tool configuration:
   - **Name:**`get_user_tier`
   - **Description:** "Retrieves subscription tier and last login for a user given their email."
   - **Parameters:**`email` (string)

### Step 5: Publish & Connect

1. Click **"Publish"** in the right sidebar.
2. Click **"Generate Token"**.
3. Copy your **MCP HTTP Stream URL** (`https://mcp.publish.pylar.ai/mcp`) and **Authorization Bearer Token**.

**Connect to Vercel AI SDK:**
If you're building a Next.js app with Supabase, you can use Pylar tools in your AI route.

```typescript
// app/api/chat/route.ts
import { MCPTool } from "langchain-mcp";

const pylarTool = new MCPTool({
  url: "https://api.pylar.ai/mcp/v1/server/YOUR_SERVER_ID",
  apiKey: process.env.PYLAR_API_KEY,
});

// Pass this tool to your Vercel AI SDK stream
```

## Advanced Use Cases for Supabase Agents

### 1. Content Moderation Bot

**Goal:** An agent that scans new comments and flags them, without giving it write access to the whole DB.
- **View:**`comments` table filtered by `created_at > NOW() - INTERVAL '1 hour'.`
- **Tool Prompt:** "Create a tool that fetches recent comments for moderation."
- **User Query:** "Check the last hour of comments for spam."

### 2. SaaS Onboarding Assistant

**Goal:** An in-app assistant that helps users set up their account based on their current progress.
- **View:**`onboarding_steps` joined with `user_profiles`.
- **Tool Prompt:** "Create a tool that checks which onboarding steps a user has completed."
- **User Query:** "What should I do next to finish setting up my account?"

## Conclusion

Building AI features on top of Supabase doesn't have to mean compromising on security or spending weeks writing boilerplate code. With Pylar, you can create a secure, governed MCP server for your Supabase project in under 2 minutes—no coding required.

By connecting Pylar to your Supabase Postgres instance, creating sandboxed views that protect sensitive data, and auto-generating MCP tools with natural language, you unlock the full potential of your Supabase data for AI agents while maintaining complete control and security.

Whether you're building a content moderation bot, a SaaS onboarding assistant, or any other AI-powered feature, Pylar gives you the governance layer you need to deploy agents safely. Your `service_role` keys stay secure, your data stays protected, and your agents get exactly the access they need—nothing more.