MCP Transport: stdio vs SSE

Unicon MCP supports two connection methods. This guide helps you choose the right one.

Quick Decision

Use the app you're using to decide:

Your App	Transport	Why
Claude Desktop	stdio	Only supports stdio
Claude Code (CLI)	stdio	Local Node.js available
Cursor IDE	Either	Supports both, stdio recommended
v0 (Vercel)	SSE	Cloud-based, no local process
Bolt / Lovable	SSE	Cloud-based, no local process
Replit	SSE	Cloud-based, no local process
Custom web app	SSE	Browser environment

TL;DR: Desktop apps → stdio. Cloud/browser apps → SSE.

What's the Difference?

stdio (Standard I/O)

A local Node.js process runs on your machine. Your AI app talks to it via pipes.

Your App ←→ Local Process (npx) ←→ Unicon API

Requires: Node.js/npm installed locally

SSE (Server-Sent Events / HTTP)

Direct HTTP connection to our hosted endpoint. Nothing runs locally.

Your App ←→ unicon.sh/api/mcp

Requires: Network access only

Comparison

Factor	stdio	SSE
Setup complexity	Config file + restart	Just paste URL
Local dependencies	Needs Node.js	None
Latency	Lower (local process)	Slightly higher (HTTP)
Offline capability	Partial (process cached)	None
Works in browser	No	Yes
Resource usage	Spawns background process	None
Always latest version	npm update needed	Automatic

The Bottom Line

Both give you the exact same tools and results. The only difference is how your app connects to Unicon.

If your app supports both, stdio is marginally faster but SSE is simpler to set up. For most users, the difference is imperceptible.

SSE Endpoint URL

MCP Endpoint

https://unicon.sh/api/mcp

This is a Streamable HTTP endpoint that supports both SSE and standard HTTP responses.

How It Works

┌─────────────────────────────┐
│   v0 / Cloud IDE / Agent    │
└──────────┬──────────────────┘
           │ MCP Protocol (Streamable HTTP)
           │
┌──────────▼──────────────────┐
│ unicon.sh/api/mcp │
└─────────────────────────────┘

The endpoint handles MCP protocol messages directly over HTTP. No local installation required.

Configuration Examples

v0 (Vercel)

Add Unicon as an MCP server in v0's settings:

https://unicon.sh/api/mcp

Bolt / Lovable

Configure in project settings:

mcp.json

{
  "mcpServers": {
    "unicon": {
      "url": "https://unicon.sh/api/mcp"
    }
  }
}

Custom Integration

For custom MCP clients using the SDK:

TypeScript

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StreamableHTTPClientTransport } from "@modelcontextprotocol/sdk/client/streamableHttp.js";

const transport = new StreamableHTTPClientTransport(
  new URL("https://unicon.sh/api/mcp")
);

const client = new Client({
  name: "my-app",
  version: "1.0.0"
}, {
  capabilities: {}
});

await client.connect(transport);

// Now you can call tools
const result = await client.callTool({
  name: "search_icons",
  arguments: { query: "dashboard", limit: 10 }
});

Available Tools

The SSE endpoint provides the same tools as the stdio transport:

Tool	Description
search_icons	AI-powered semantic search
get_icon	Get single icon by ID
get_multiple_icons	Get multiple icons at once
list_libraries	List available icon sources
list_categories	List icon categories
get_starter_pack	Get curated icon collections

CORS Support

The endpoint supports CORS for browser-based integrations:

Access-Control-Allow-Origin: *
Access-Control-Allow-Methods: GET, POST, DELETE, OPTIONS
Access-Control-Allow-Headers: Content-Type, Accept, Mcp-Session-Id, Mcp-Protocol-Version, Last-Event-Id

Rate Limits

Tier	Requests	Window
Free	100	per minute
Burst	10	per second

Troubleshooting

Connection Issues

1. Verify the URL is exactly https://unicon.sh/api/mcp
2. Check that your client supports Streamable HTTP transport
3. Ensure network allows outbound HTTPS connections

Slow Responses

First requests may be slower due to cold starts. Subsequent requests are fast due to edge caching.

Tool Not Found

Verify you're using the correct tool names (search_icons, get_icon, etc.)