Perplexity MCP Server

What It Does

Four specialized AI-powered research tools from Perplexity: web search (quick answers with citations), chat (conversational AI with real-time web access via sonar-pro), deep research (comprehensive multi-step research via sonar-deep-research), and reason (advanced reasoning via sonar-reasoning-pro with showThinking parameter).

Each tool targets a different research depth — from fast factual lookups to thorough investigative research that takes 30+ seconds to complete.

What It Does Well

• Four tools at different depth levels is the right architecture. Quick search for facts, chat for conversational follow-ups, deep research for comprehensive analysis, and reason for complex problem-solving. Claude picks the right tool for the query complexity instead of running every search through the same one-size-fits-all pipeline.
• Citations and source filtering build trust in results. Every response includes citations to original sources. Recency filters (day/week/month/year), domain filtering, and academic mode let you control where information comes from. This is measurably better than search MCP servers that return results without provenance.
• Deep research produces report-quality output. The sonar-deep-research model runs multi-step research that takes 30+ seconds but returns comprehensive, sourced analysis. For tasks where Claude needs to understand a topic thoroughly before acting, this replaces manual multi-query research workflows.

What It Doesn’t Do Well

• Requires a Perplexity API key, which is paid. Unlike Brave Search MCP (free, no key) or Fetch MCP (free, no key), Perplexity requires a paid API subscription. The quality is higher, but so is the barrier to entry.
• Deep research timeout needs manual extension in some clients. The sonar-deep-research model can take 180+ seconds. Default MCP timeouts in some clients will kill the request before it completes. You may need to configure extended timeouts, which isn’t always straightforward.

Setup Notes

Standard npx install with API key. Get the key from perplexity.ai/settings/api. For Claude Code: claude mcp add perplexity --env PERPLEXITY_API_KEY="pplx-your_key" -- npx -y @perplexity-ai/mcp-server.

HTTP mode is available via npm run start:http for remote setups, exposing the server at http://localhost:8080/mcp.

If you plan to use deep_research, set your MCP client timeout to at least 180 seconds to avoid premature request termination.

Config

{
  "mcpServers": {
    "perplexity": {
      "command": "npx",
      "args": ["-y", "@perplexity-ai/mcp-server"],
      "env": {
        "PERPLEXITY_API_KEY": "pplx-your_api_key"
      }
    }
  }
}

Tested With

• Claude Desktop on Windows 11

{
  "mcpServers": {
    "perplexity-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@perplexity-ai/mcp-server"
      ],
      "env": {
        "PERPLEXITY_API_KEY": "pplx-your_api_key"
      }
    }
  }
}