La integración MCP de Perplexity cuesta entre 5 y 14 dólares por cada 1.000 solicitudes más tarifas por token. Puede crear una capacidad similar utilizando una API de búsqueda estructurada: búsqueda en múltiples fuentes con URL de citas que Claude puede sintetizar en respuestas citadas. Este tutorial construye ese patrón utilizando el servidor MCP de Scavio.
Requisitos previos
- Claude Desktop o Claude Code con soporte MCP
- Una clave API de Scavio
Guia paso a paso
Paso 1: Conecte Scavio MCP a Claude
Agregue el servidor MCP alojado de Scavio a su configuración de Claude Desktop o Claude Code.
JSON
// For Claude Desktop, add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"scavio": {
"url": "https://mcp.scavio.dev/mcp",
"headers": {
"x-api-key": "your_scavio_api_key"
}
}
}
}
// For Claude Code, add to .mcp.json in your project:
{
"mcpServers": {
"scavio": {
"url": "https://mcp.scavio.dev/mcp",
"headers": { "x-api-key": "your_scavio_api_key" }
}
}
}Paso 2: Utilice la búsqueda de múltiples fuentes para las respuestas citadas
Pídale a Claude que busque en múltiples plataformas y cite fuentes en su respuesta.
Bash
# In Claude Desktop/Code, you can now ask:
# 'Search Google and Reddit for the current state of the SerpAPI lawsuit
# and summarize with citations.'
# Claude will:
# 1. Call google_search('serpapi google lawsuit 2026')
# 2. Call reddit_search('serpapi lawsuit update')
# 3. Synthesize results with [Source](URL) citations
# This gives Perplexity-style cited answers without per-token API fees.
# Cost: 2 credits ($0.01) vs Perplexity Sonar ($0.005-0.014 per request + tokens)Paso 3: Cree un mensaje de búsqueda múltiple personalizado
Cree un patrón de aviso del sistema que le indique a Claude que busque múltiples fuentes y las cite.
Python
# Save this as a Claude Code custom command or system prompt:
RESEARCH_PROMPT = """
When answering research questions:
1. Search Google for factual/official sources
2. Search Reddit for community opinions and experiences
3. Search YouTube for relevant video content (if applicable)
4. Synthesize findings into a clear answer
5. Cite each claim with [Source Title](URL)
6. Note any conflicting information between sources
Always search at least 2 platforms before answering.
"""Paso 4: Compare costos con Perplexity Sonar
Calcule la diferencia de costo entre este enfoque y la API de Perplexity.
Python
# Perplexity Sonar API pricing (verified May 2026):
# - Base model: $1/$1 per million tokens + $5/1K requests
# - Pro model: $3/$15 per million tokens + $14/1K requests
# Example: 100 research queries per day
perplexity_base = (100 * 5 / 1000) + (100 * 500 * 2 / 1_000_000) # ~$0.60/day
perplexity_pro = (100 * 14 / 1000) + (100 * 500 * 18 / 1_000_000) # ~$2.30/day
# Scavio MCP approach: 2-3 searches per query * 100 queries = 200-300 credits
scavio_cost = 300 * 0.005 # $1.50/day
# But Claude is already paid for (subscription). Only additional cost is search API.
# If on free tier: 250 free credits covers 166 multi-source queries/month
print(f'Perplexity Base: ${perplexity_base:.2f}/day for 100 queries')
print(f'Perplexity Pro: ${perplexity_pro:.2f}/day for 100 queries')
print(f'Scavio MCP: ${scavio_cost:.2f}/day for 100 queries (search cost only)')Ejemplo en Python
Python
# No Python SDK needed - MCP handles the integration.
# Scavio MCP server URL: https://mcp.scavio.dev/mcp
# Claude automatically discovers: google_search, reddit_search, youtube_search,
# amazon_search, walmart_search tools when connected.Ejemplo en JavaScript
JavaScript
// MCP configuration is JSON-based, no code needed:
// Add to your MCP client config:
const mcpConfig = {
scavio: {
url: 'https://mcp.scavio.dev/mcp',
headers: { 'x-api-key': process.env.SCAVIO_API_KEY }
}
};Salida esperada
JSON
A Perplexity-style research capability in Claude Desktop/Code using Scavio MCP, providing multi-source cited answers at lower cost than Perplexity Sonar API.