Answers, not links — delegate questions to Perplexity's search-grounded models and get cited responses back.
Most search MCP servers hand your assistant raw material — links, snippets, page content — and leave the reading to it. Perplexity's server inverts that: the assistant delegates the whole question to Perplexity's Sonar models, which run their own live web searches, read the results, and return a finished answer with citations. Search-as-a-colleague rather than search-as-a-library.
That's an efficient trade for questions where you want a current, synthesized take: "what changed in the latest Node LTS?", "what's the consensus on this drug trial?", "who won and what did commentators make of it?". One tool call replaces a search-fetch-read-summarise loop, and the citations keep the answer checkable — the non-negotiable feature for delegated research.
The server (Anthropic-listed, maintained in Perplexity's modelcontextprotocol repo) exposes the Sonar family tiers as tools: quick ask for everyday questions, a research variant for deeper multi-source reports, and a reasoning option for questions that need thinking as well as finding. Cost scales with tier, billed through your Perplexity API account.
The philosophical trade-off is worth being honest about: you're trusting Perplexity's reading of the web rather than doing your own. For quick current-events grounding, that trade is excellent. For research where you want to choose the sources, a raw search server (Brave, Exa) keeps the judgment on your side. Plenty of setups run both and route by question type — a pattern that works well in practice.
messages array, not a bare string. Passing the surrounding conversation lets Sonar build much better search queries than a decontextualised question ever produces.day, week, month) is the difference between "what changed this week" and a confident, well-written summary of last year's state of play.sonar-deep-research bills reasoning tokens and long citation-heavy output on top of the search itself, so route to it deliberately; a routine ask costs a small fraction of one deep research run.One call turns 'what happened with X this week' into a sourced, current answer.
Hand a research question off and get a multi-source report with citations.
Verify claims against the live web mid-conversation.
| Tool | What it does |
|---|---|
| perplexity_ask | Ask a question; Sonar searches the web and returns a cited answer |
| perplexity_research | Deeper multi-source research via sonar-deep-research |
| perplexity_reason | Reasoning-heavy questions via sonar-reasoning-pro |
{
"mcpServers": {
"perplexity": {
"command": "npx",
"args": ["-y", "server-perplexity-ask"],
"env": { "PERPLEXITY_API_KEY": "pplx-your-key" }
}
}
}API keys are created in Perplexity account settings (API section).
A Perplexity API key with pay-as-you-go billing. Research and reasoning tiers cost more per call than quick asks.
| Variable | Description | Required |
|---|---|---|
| PERPLEXITY_API_KEY | Perplexity API key (pay-as-you-go billing)e.g. pplx-... | Yes |
The simplest web tool that matters — fetch any URL and get model-ready markdown back.
Industrial-strength web extraction — render, scrape, crawl and search entire sites into clean markdown.
Puppeteer-powered browser control that drives pages from the accessibility tree instead of pixels.
Search built for AI, not humans — semantic web search that returns model-ready content, plus code context.
Put 6,000+ pre-built scrapers at your assistant's fingertips through one MCP endpoint.
Independent-index web search for assistants — fresh results without the Google API maze.