Use the simple ask for a lookup and the chat tool when the follow-up depends on what was said before — passing the message history is what keeps a multi-turn research thread coherent. Model choice is per call, so a cheap small model can handle routine lookups while the large one is reserved for the hard questions.
A small Node.js bridge between an MCP client and the Perplexity API. It exists so a model that cannot search can hand the question to one that can, and get back an answer grounded in current web results.
- `perplexity_ask` — send one `query` and get the answer back, with an optional `model`
- `perplexity_chat` — a full chat completion: `messages` as an array of role and content objects, plus optional `model` and a `temperature` between 0 and 2
- A choice of three online models on either tool: `llama-3.1-sonar-small-128k-online`, `llama-3.1-sonar-large-128k-online`, `llama-3.1-sonar-huge-128k-online`
A Perplexity API key, set as `PERPLEXITY_API_KEY` in the client's env block. There is no published binary: clone the repository, `npm install`, `npm run build`, then point the config at `node path/to/perplexity-mcp-server/dist/index.js`. Communicates over stdio. Package `perplexity-mcp-server` at 1.0.0, ISC licensed.
Build from source — clone the repository and build it, then point your client at the binary