A single well-named tool costs almost nothing in context and needs no explanation to the model — it either asks Perplexity or it does not. If what you want is a sourced web answer inside a conversation, this is the smallest thing that does it. If you want document extraction, persistent threads or developer-shaped research tools, this is not that.
A small MCP server over the Perplexity API. It does one thing — `ask_perplexity` requests a chat completion and returns it with citations — and it installs in a single line.
- `ask_perplexity` — a chat completion with the citations Perplexity used, so an answer arrives with sources attached
- `uvx mcp-server-perplexity` with the key in the environment: nothing to clone, nothing to build
A Perplexity API key in `PERPLEXITY_API_KEY`, and `uv` so that `uvx` can fetch the package — `mcp-server-perplexity`. One documented limitation to plan around: the server does not report progress on long-running calls, so Claude Desktop can time out when Perplexity takes a while. Keep questions tight, and expect the occasional dropped answer on a heavy one.
One command plus a key — uvx mcp-server-perplexity, then supply credentials