Skyfire MCP in Action

This example demonstrates how easily an agent can orchestrate multiple MCP servers in a single workflow to conduct agentic commerce.

Using production Skyfire and Dappier MCP servers, the agent:

Connects to the Skyfire MCP server to create a kya token
Uses that token to authenticate with Dappier’s MCP server
Purchases and retrieves real-time data

With a Skyfire API key, Dappier API key, and OpenAI API key, run this demo in one click to see a real-time data purchase.

import asyncio
import os

from agents import Agent, Runner
from agents.mcp import MCPServerStreamableHttp

# Define MCP servers
skyfire_server = MCPServerStreamableHttp(
    name="skyfire",
    params={
        "url": "https://mcp.skyfire.xyz/mcp",
        "headers": {
            "skyfire-api-key": os.getenv("SKYFIRE_API_KEY"),
        },
    },
)

dappier_server = MCPServerStreamableHttp(
    name="dappier",
    params={
        "url": f"https://mcp.dappier.com/mcp?apiKey={os.getenv('DAPPIER_API_KEY')}",
    },
)

# Define agent
agent = Agent(
    name="General purpose agent",
    mcp_servers=[skyfire_server, dappier_server],
    mcp_config={
        "convert_schemas_to_strict": True,
        "failure_error_function": None,
    },
)


async def main():
# Define agent prompt
    prompt = """
You have access to MCP tools from two servers: skyfire and dappier.

Step 1: Call the Skyfire tool that creates a KYA token.
Step 2: Using the Skyfire token, call the real-time-search Dappier tool.
- query: "Latest AI news today, include date and time"
- pass the Skyfire token as the auth credential per the tool schema.
"""
    print("\nConnecting to MCP servers (Skyfire, Dappier)...\n")
    async with skyfire_server, dappier_server:
        result = await Runner.run(agent, prompt)
        print("=== Agent run completed ===")
        print(result.final_output)

if __name__ == "__main__":
    asyncio.run(main())

After running the script, visit the Skyfire Dashboard.

You will see:

A newly created kya token issued for the Dappier Search Seller Service
The token linked to the Seller Service ID used in the demo

This confirms that:

The agent successfully authenticated via Skyfire
A real identity-enabled service interaction occurred

You can inspect the token details directly in the Dashboard for more information.

Low-Level MCP Operations

If you prefer to interact directly with the Skyfire MCP server via JSON-RPC, here are the core operations.

1. Connect

Initialize a session:

curl -i https://mcp.skyfire.xyz/mcp \
  -H "Content-Type: application/json" \
  -H "skyfire-api-key: <BUYER_AGENT_API_KEY>" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": "101",
    "method": "initialize",
    "params": {
      "protocolVersion": "123",
      "capabilities": {},
      "clientInfo": {
        "name": "curl",
        "version": "1.0"
      }
    }
  }'

Extract the mcp-session-id header from the response. Use this header for all subsequent calls:

mcp-session-id: <SESSION_ID>

2. List tools (Optional)

curl -X POST https://mcp.skyfire.xyz/mcp \
  -H "Content-Type: application/json" \
  -H "skyfire-api-key: <BUYER_AGENT_API_KEY>" \
  -H "mcp-session-id: <SESSION_ID>" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": "103",
    "method": "tools/list",
    "params": {}
  }'

3. Call tools

find-sellers

curl -X POST https://mcp.skyfire.xyz/mcp \
  -H "Content-Type: application/json" \
  -H "skyfire-api-key: <BUYER_AGENT_API_KEY>" \
  -H "mcp-session-id: <SESSION_ID>" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": "104",
    "method": "tools/call",
    "params": {
      "name": "find-sellers",
      "arguments": { "search": "" }
    }
  }'

create-kya-token

curl -X POST https://mcp.skyfire.xyz/mcp \
  -H "Content-Type: application/json" \
  -H "skyfire-api-key: <BUYER_AGENT_API_KEY>" \
  -H "mcp-session-id: <SESSION_ID>" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": "105",
    "method": "tools/call",
    "params": {
      "name": "create-kya-token",
      "arguments": {
        "sellerServiceId": "<SELLER SERVICE ID>"
      }
    }
  }'

create-pay-token

curl -X POST https://mcp.skyfire.xyz/mcp \
  -H "Content-Type: application/json" \
  -H "skyfire-api-key: <BUYER_AGENT_API_KEY>" \
  -H "mcp-session-id: <SESSION_ID>" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": "106",
    "method": "tools/call",
    "params": {
      "name": "create-pay-token",
      "arguments": {
        "amount": "<AMOUNT_IN_USD>",
        "sellerServiceId": "<SELLER_SERVICE_ID>"
      }
    }
  }'

create-kya-payment-token

curl -X POST https://mcp.skyfire.xyz/mcp \
  -H "Content-Type: application/json" \
  -H "skyfire-api-key: <BUYER_AGENT_API_KEY>" \
  -H "mcp-session-id: <SESSION_ID>" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": "106",
    "method": "tools/call",
    "params": {
      "name": "create-kya-payment-token",
      "arguments": {
        "amount": "<AMOUNT_IN_USD>",
        "sellerServiceId": "<BUYER_AGENT_API_KEY>"
      }
    }
  }'

Connect to the Skyfire MCP server from Claude Desktop

Open claude_desktop_config.json from Settings and add the following configuration.

{
    "mcpServers": {
        "skyfire": {
            "command": "npx",
            "args": [
                "-y",
                "mcp-remote",
                "https://mcp.skyfire.xyz/mcp",
                "--header",
                "skyfire-api-key: <BUYER_AGENT_API_KEY>"
            ]
        }
    }
}

Restart Claude Desktop.