This is a thin bridge, and the useful knob is `PERPLEXITY_MODEL` — the difference between `sonar` and `sonar-reasoning-pro` is a real difference in what you get back and what it costs. Since search is invoked at the model's discretion, keep the "Search the web" prefix in mind for when it decides it already knows.
A small MCP server that hands web search to Perplexity's API. Ask a question that needs current information and the assistant calls out to Perplexity and brings the answer back into the conversation.
- Web search results returned in the conversation, so questions about recent events or specific places get answered from the live web
- Model selection through `PERPLEXITY_MODEL`: `sonar-reasoning-pro` for the most capable reasoning, `sonar-reasoning` for enhanced reasoning, `sonar-pro` for faster responses, and `sonar` as the default
- A manual trigger — prefix a prompt with "Search the web" when the assistant does not reach for it on its own
A Perplexity API key from your Perplexity account settings, set as `PERPLEXITY_API_KEY`. Node.js v14 or higher. Install through Smithery with `npx -y @smithery/cli install @jaacob/perplexity-mcp --client claude`, or clone the repository, `npm install`, `npm run build`, and point the config at `build/index.js`. The package is `perplexity-server` 1.0.0, MIT licensed. Perplexity bills you for usage at its own published rates.
One command plus a key — npx -y @smithery/cli install @jaacob/perplexity-mcp --client claude, then supply credentials