Skip to main content
Connections give the AI access to your data and tools. This guide covers adding an MCP server, the most common connection type.

Prerequisites

  • A running MCP server (HTTP or STDIO)
  • A workspace to add the connection to

Add via dashboard

  1. Navigate to your workspace
  2. Open the Connections tab
  3. Click Add connection
  4. Choose the connection type (MCP)
  5. Fill in the connection details:

HTTP transport

FieldDescriptionExample
NameDisplay name for this connectionorder-service
URLThe MCP server’s HTTP endpointhttp://localhost:3000/mcp

STDIO transport

FieldDescriptionExample
NameDisplay name for this connectionorder-service
CommandThe executable to runnode
ArgsCommand arguments["/opt/services/order-mcp/index.js"]
  1. Click Test to verify the connection and discover tools
  2. Click Save to add the connection

Add via API

Test the connection first

curl -X POST https://your-instance/api/connectors/mcp/test \
  -H "Content-Type: application/json" \
  -H "Cookie: supaproxy_session=YOUR_JWT" \
  -d '{
    "transport": "http",
    "url": "http://localhost:3000/mcp"
  }'
Response:
{
  "ok": true,
  "tools": 4,
  "server": "order-service",
  "toolNames": ["get_order", "list_orders", "create_order", "update_order"]
}

Save the connection

curl -X POST https://your-instance/api/connectors/mcp \
  -H "Content-Type: application/json" \
  -H "Cookie: supaproxy_session=YOUR_JWT" \
  -d '{
    "workspace_id": "ws-example",
    "name": "order-service",
    "transport": "http",
    "url": "http://localhost:3000/mcp"
  }'

Add via SDK

// Test
const test = await client.connectors.testMcp({
  transport: 'http',
  url: 'http://localhost:3000/mcp'
});

if (test.ok) {
  // Save
  await client.connectors.addMcp({
    workspace_id: 'ws-example',
    name: 'order-service',
    transport: 'http',
    url: 'http://localhost:3000/mcp'
  });
}

Tool discovery

For HTTP connections, tools are discovered automatically when the connection is saved. For STDIO connections, tools are discovered on the first query. Each discovered tool includes:
  • Name: unique identifier
  • Description: what the tool does
  • Input schema: JSON Schema for parameters
  • Is write: whether the tool modifies data
Write tools are flagged and subject to the write confirmation guardrail if enabled.