← Connecting AI Agents | Next: Complete Source Code Reference →

Current Limitations

As of this writing, Supabase Edge Functions MCP servers use --no-verify-jwt, meaning the MCP endpoint is publicly accessible. This is fine for development and internal tools, but production systems should add authentication.

Adding Basic API Key Authentication

You can add a simple API key check inside your MCP server:

// Add this middleware before the MCP handler
app.use('*', async (c, next) => {
  const apiKey = c.req.header('X-API-Key')
  const expectedKey = Deno.env.get('MCP_API_KEY')
 
  if (expectedKey && apiKey !== expectedKey) {
    return c.json({ error: 'Unauthorized' }, 401)
  }
 
  await next()
})

Set the secret:

supabase secrets set MCP_API_KEY=your-secret-key-here

Then configure your MCP client to send the header (varies by client).

Best Practices

  1. Use the service role key carefully: The service role key bypasses Row-Level Security. Only use it when your MCP server needs unrestricted access. For multi-tenant scenarios, pass user tokens and use the anon key with RLS policies.

  2. Validate inputs: Zod schemas handle type validation, but add business logic validation in your handlers (e.g., checking that a category exists before creating a product).

  3. Limit result sizes: Always use .limit() on database queries to prevent returning unbounded result sets.

  4. Log operations: Add logging for audit trails:

    console.log(`[MCP] Tool called: ${toolName}`, args)

  5. Destructive operations need confirmation: Note how our delete_product tool requires a confirm: true parameter. Apply this pattern to any destructive operation.

  6. Write descriptive tool descriptions: The AI agent reads your description field to decide when and how to use each tool. Vague descriptions lead to misuse. Be specific about what the tool does, what it returns, and when to use it.

  7. Return structured data alongside formatted text: Return both human-readable summaries and raw JSON, so the AI can both present information to users and process it programmatically.

  8. Handle errors gracefully: Always catch errors and return meaningful error messages via isError: true. This helps the AI agent understand what went wrong and potentially retry with different parameters.

Next: Module 13 - Complete Source Code Reference →