Provides access to Perplexity's AI models through two tools: 'perplexity_ask' for sending single questions and 'perplexity_chat' for multi-turn conversations, with support for customizable model selection.
Perplexity Intelligent MCP Server
An intelligent Perplexity research MCP server for Claude Desktop with automatic model selection, multi-model synthesis, and persistent research history.
Features
Intelligent Model Selection: Auto-picks the right Perplexity model based on your query
Research History: Saves all research to
~/Documents/Perplexity Research/Multi-Model Synthesis: Combines multiple models for comprehensive research
Search History: Search past research to avoid duplicate queries
Built-in Citations: Every response includes sources and references
Related MCP server: Perplexity MCP Server
Quick Setup
Prerequisites
Node.js 18+ (download)
Claude Desktop
Perplexity API key (get one at perplexity.ai/settings/api)
Configure Claude Desktop
Open ~/Library/Application Support/Claude/claude_desktop_config.json and add:
Replace your-api-key-here with your Perplexity API key.
Restart Claude Desktop
Quit completely (Cmd+Q) and reopen. You're done!
Available Tools
research
Intelligent Perplexity search with auto-logging.
deep_research
Multi-model synthesis for comprehensive research.
search_research
Search past research for existing findings.
list_research_threads
List all saved research threads.
read_research_thread
Read a specific past research thread.
list_models
Show available Perplexity models and when to use each.
Models
Model | Best For |
sonar | Simple factual lookups, current events |
sonar-pro | Multi-source research, fact-checking |
sonar-reasoning-pro | Why/how questions, complex analysis (default) |
sonar-deep-research | Exhaustive research, comprehensive reports |
Research Storage
All research is automatically saved to:
You can browse your research history in Finder anytime.
Environment Variables
Variable | Description | Default |
| Your Perplexity API key | Required |
| Custom research directory |
|
License
MIT