Perplexity MCP Server

What is Perplexity MCP Server?

An MCP (Model Context Protocol) server providing comprehensive access to the Perplexity AI API. It enables LLMs and AI agents to perform fast, search-augmented queries and conduct exhaustive, multi-source deep research through a standardized, secure, and easy-to-integrate protocol. It is built for developers using MCP clients like advanced AI assistants, IDE extensions, or custom research tools.

How to use Perplexity MCP Server?

Install Node.js ≥18 and obtain a Perplexity API key. Clone the repository, run npm install and npm run build. Create a .env file with PERPLEXITY_API_KEY. Configure your MCP client (e.g., Cline) to run the server with the command node /path/to/dist/index.js and pass the environment variable. Start the server with npm start (uses stdio by default) or configure HTTP transport.

Key features of Perplexity MCP Server

• Dual API support for quick search and deep research
• Advanced search control (recency, domain, academic mode)
• Cost tracking utility based on token usage
• Structured logging, centralized error handling, input validation
• HTTP transport with JWT and OAuth 2.1 authentication
• Resilient API client with error handling and logging

Use cases of Perplexity MCP Server

• Automate research by programmatically performing quick lookups or deep-dive investigations
• Enhance AI reasoning by grounding LLM responses with up-to-date, verifiable web information
• Integrate search‑augmented generation into any AI‑driven workflow
• Generate detailed multi‑source reports on complex topics

FAQ from Perplexity MCP Server

What are the prerequisites to run Perplexity MCP Server?

Node.js version 18 or higher, npm, and a Perplexity API key (available at perplexity.ai/settings/api).

How do I configure the server?

Create a .env file with PERPLEXITY_API_KEY (required). Optionally set MCP_TRANSPORT_TYPE (stdio or http), MCP_HTTP_PORT, MCP_AUTH_MODE, and other environment variables documented in the README.

What transport mechanisms are supported?

stdio (default) and HTTP. HTTP can be secured with JWT or OAuth 2.1 authentication.

Is there a cost tracking feature?

Yes, the server includes a utility to estimate API call costs based on token usage and model, helping manage expenses.

What timeout should I use for the deep research tool?

The deep research tool may take longer than 60 seconds; the README recommends allowing a timeout of 180 seconds in MCP clients like Cline.