@perplexity-ai/mcp-server

What it does

The official MCP server for the Perplexity API Platform, exposing four tools for AI assistants: perplexity_search for ranked web search results, perplexity_ask for conversational queries with real-time context, perplexity_research for deep analysis using the sonar-deep-research model, and perplexity_reason for complex reasoning tasks with the sonar-reasoning-pro model. All tools ground responses in current web data via Perplexity's Sonar APIs.

Who it's for

Claude or Cursor users who need their AI assistant to search the live web and provide current, fact-checked answers without hallucination. Useful for researchers, developers debugging third-party APIs, analysts tracking news or market data, and anyone asking questions that require real-time context.

Common use cases

• Query current events, market data, or recent news without stale training data.
• Cross-check or verify claims by grounding responses in real web sources.
• Deep-dive research on a topic with the sonar-deep-research model for reports or documentation.
• Solve complex reasoning problems that benefit from access to current information.
• Use with Cursor or Claude Code to debug APIs and services by pulling their latest docs and status pages.

Setup pitfalls

• PERPLEXITY_API_KEY must be set; requests fail silently if the env var is missing or empty.
• Timeout defaults to 5 minutes; research and reasoning queries may exceed this—set PERPLEXITY_TIMEOUT_MS=600000 or higher if you see timeout errors.
• If behind a corporate proxy, set PERPLEXITY_PROXY=https://your-proxy-host:8080; requests will hang if proxy is needed but not configured.
• The strip_thinking parameter only applies to perplexity_reason and perplexity_research; using it on other tools has no effect.