Official Search & Web

Perplexity

Search-grounded answers powered by Perplexity's API.

Works with: Claude DesktopClaude CodeCursor
Quick install
npx -y @perplexity/mcp-server

How to install the Perplexity MCP server

Add this to your Claude Desktop MCP configuration:

{
  "mcpServers": {
    "perplexity": {
      "command": "npx",
      "args": [
        "-y",
        "@perplexity/mcp-server"
      ]
    }
  }
}

Add this to your Claude Code MCP configuration:

npx -y @perplexity/mcp-server

Add this to your Cursor MCP configuration:

{
  "mcpServers": {
    "perplexity": {
      "command": "npx",
      "args": [
        "-y",
        "@perplexity/mcp-server"
      ]
    }
  }
}

The Perplexity MCP server gives Claude access to Perplexity’s search-grounded answer API. Where Brave and Tavily return search results for Claude to synthesise, Perplexity returns a synthesised answer with inline citations. For quick factual questions, this is the lower-token, higher-quality option.

Why use it

Perplexity’s whole product is “search and synthesise in one step”. Wrapping it as an MCP server gives Claude that capability natively. Ask a factual question, get an answer with sources, no manual synthesis required.

For workflows where Claude needs to ground a single sentence (“the EU AI Act came into force in…”) rather than do deep research, Perplexity is faster and cheaper than running search + fetch + synthesis.

What it actually does

Send a query, optionally pick a model and search domain (web, academic, news), receive a synthesised answer with citations. Some servers also expose chat-style follow-up.

Practical patterns:

  • “What’s the current Bank of England base rate?”
  • “Has Cloudflare announced anything new about Workers AI in the last week?”
  • “What did the Anthropic team say at the most recent dev event about MCP?”

Gotchas

API costs scale with model tier. Sonar Pro is significantly more expensive than Sonar. Default to Sonar for simple queries; only escalate when needed.

Citations are not always perfectly accurate. Perplexity can hallucinate citations occasionally. For high-stakes claims, verify by reading the cited source.

Pair with Tavily for “give me a grounded answer + raw sources to back it up”. Claude composes the workflow automatically.

Perplexity MCP server: FAQs

How is this different from running Brave or Tavily plus Fetch?

Perplexity's API returns synthesised answers with inline citations rather than raw search results. Claude consumes a single grounded answer instead of multiple sources. Better for quick factual questions, weaker for research where you want to read sources.

What does the API cost?

Perplexity's Sonar API has multiple pricing tiers. Free tier exists with rate limits; paid tiers scale up. Check perplexity.ai/api for current pricing.

Can I pick which Perplexity model to use?

Yes. The server exposes model selection (Sonar, Sonar Pro, etc). Sonar is cheap and fast; Pro is better for complex queries.

Does it support follow-up queries?

Each MCP call is independent, but Claude can issue multiple calls naturally as a conversation progresses. Perplexity treats each query in isolation.

Should I use Perplexity instead of Brave Search?

Use both. Perplexity for 'give me a quick grounded answer'. Brave or Tavily for 'find me sources I can read'. Claude routes between them based on the prompt.