The official Perplexity "Ask" MCP server lets an AI client run live, citation-backed web queries through Perplexity's Sonar API — so your agent can ground answers in current information. Here's the setup.
Step 1 — get a Perplexity API key
In your Perplexity account, open the API settings and create a key (Sonar API). Usage is pay-as-you-go.
Step 2 — configure the client
{
"mcpServers": {
"perplexity": {
"command": "npx",
"args": ["-y", "server-perplexity-ask"],
"env": {
"PERPLEXITY_API_KEY": "your_api_key"
}
}
}
}
Restart the client; see MCP config file location.
What it gives your AI
A perplexity_ask tool that performs a real web search and returns a synthesized, cited answer — useful for current events, docs and research where the model's training data is stale.
Verify
Ask: "Use Perplexity to find the newest MCP features announced this month, with sources." Fresh, cited results back confirm it works.
Common problems
401 unauthorized— wrong API key, or no Sonar API access on the account.- Empty/limited answers — check your API credit balance.
- No tools in client — Node 18+ for
npx; see Cursor MCP not working.
Perplexity vs other search servers
Perplexity returns a synthesized, cited answer; Tavily and Bright Data lean toward raw results/extraction you process yourself. Many stacks use one search server — pick by whether you want answers or raw data.
Going further
Pair with Sequential Thinking and a filesystem server for research workflows. Browse the search category or curated loadouts. Keep the key private — see MCP security best practices.