What it does
This is Perplexity's official MCP server, exposing four tools for real-time web search and AI reasoning. perplexity_search queries the Search API directly for ranked results; perplexity_ask handles conversational Q&A with web search context via the sonar-pro model; perplexity_research runs deep, comprehensive research using sonar-deep-research; and perplexity_reason tackles complex analytical tasks with sonar-reasoning-pro. All tools integrate current web data into their responses, avoiding the knowledge-cutoff problem of offline models. The reasoning and research tools support an optional strip_thinking parameter to remove thinking blocks and save context.
Who it's for
Researchers, analysts, and content creators who need current information grounded in live web results. Developers building AI agents that must answer questions about recent events, market data, or breaking news. Teams using Claude or other AI editors who want their assistant to search the web rather than hallucinate outdated information.
Common use cases
- Search for recent news or current information during Claude conversations without switching tools
- Ground research reports in live web data via the deep-research model
- Solve complex problems through structured reasoning with access to current sources
- Build AI agents that answer user questions with citations to real-time web results
- Fact-check claims or find supporting evidence for analytical work
Setup pitfalls
- Requires
PERPLEXITY_API_KEYfromhttps://www.perplexity.ai/account/api/group—the server will fail silently without it - Makes live network calls to Perplexity's API; behind a corporate proxy, set
PERPLEXITY_PROXY=https://your-proxy-host:8080 - Default timeout is 5 minutes; deep research requests may approach this—increase via
PERPLEXITY_TIMEOUT_MSif requests fail - Each Sonar model (pro, deep-research, reasoning-pro) has different cost and latency profiles—test which fits your use case