Skip to content
MCP Shelf

Perplexity MCP

View on GitHub

Perplexity's search, answers and deep research, from Perplexity itself.

  • Started in our test
  • ★ 2.6K GitHub stars
  • Last commit 21 Oct 2026

At a glance

This is the official server from Perplexity. It has four tools: a plain web search that returns ranked results with dates, a quick answer tool, a reasoning tool for harder questions, and a research tool that can work for several minutes and write a detailed, cited report. The three answer tools run on Perplexity presets that the company tunes over time.

Every call goes through the Perplexity API, which has its own pay-as-you-go billing, so you need an API key with credit on it. You can run it locally with npx or point your client at the hosted copy Perplexity runs at api.perplexity.ai/mcp.

Made by
Perplexity AI
Licence
MIT
Latest version
1.3.0 (25 Sept 2026)
GitHub stars
2.6K
Last commit
21 Oct 2026
Tools
4
Runs on
Your computer or hosted

Our check: We ran this exact config on 27 Oct 2026: it started and answered with 4 tools. The shared database gave "@chatmcp/server-perplexity-ask", a package by someone else. Perplexity publishes @perplexity-ai/mcp-server, and the repository has moved to perplexityai/modelcontextprotocol.

What's in the pod: its tools

We found 4 tools when we started it on 27 Oct 2026. Every one of them only reads, so nothing changes unless you act on what it finds.

Only reads4 tools · Looks things up and changes nothing.

ToolWhat it is for
perplexity_askAnswers a question with live web search, using Perplexity's fast preset.
perplexity_reasonWorks through a harder question step by step, using the medium preset.
perplexity_researchResearches a topic in depth and writes a cited report. Can take several minutes.
perplexity_searchRuns a web search and returns ranked results with titles, links, snippets and dates.

Grouped and described by us. The README on GitHub lists the inputs each tool takes.

How clearly it describes its tools to your assistant

Tools with a description
4 of 4
Say if they change things
4 of 4
Context it takes, at least
~600 tokens

Every tool says whether it only reads or can change things, so a client can ask before the risky ones.

From the tool list it sent us. The token figure counts names, descriptions and input names; the full schemas your client loads add more.

Set it up

Pick your app for the exact snippet and where it goes.

Where it goes: claude_desktop_config.json (Settings → Developer → Edit Config)

Claude Desktop
{
  "mcpServers": {
    "perplexity": {
      "command": "npx",
      "args": [
        "-y",
        "@perplexity-ai/mcp-server"
      ],
      "env": {
        "PERPLEXITY_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

If your client fails to start it with an initialize or EOF error, use "-yq" instead of "-y" so npx stays quiet. Long research calls may need PERPLEXITY_TIMEOUT_MS raised above the 5-minute default.

Remote option

Hosted by Perplexity with the same tools. Send your API key as an "Authorization: Bearer" header.

Server URL
https://api.perplexity.ai/mcp

New to this? See the setup guides for Claude Code, Cursor, Claude Desktop and ChatGPT.

Settings

Names are exactly as the project spells them. Required ones are marked.

SettingWhat it does
PERPLEXITY_API_KEYEnvironment variable · requiredYour key from the Perplexity API console.pplx-...
PERPLEXITY_TIMEOUT_MSEnvironment variableHow long a call may run before it gives up. Defaults to five minutes.600000
PERPLEXITY_BASE_URLEnvironment variableSends requests to a different API address.https://api.perplexity.ai
PERPLEXITY_LOG_LEVELEnvironment variableDEBUG, INFO, WARN or ERROR. Defaults to ERROR.INFO
PERPLEXITY_PROXYEnvironment variableAn HTTPS proxy for corporate networks. HTTPS_PROXY and HTTP_PROXY work too.https://proxy.example.com:8080

What it can touch

It can read

  • The public web, through Perplexity's search

It can change

  • Your Perplexity API balance: each call is billed

Safer setup: Nothing it does changes your files or accounts. Set a spending limit in the API console, since research calls cost more than plain searches.

Before you start

  • A Perplexity API key with credit
  • Node.js, for the local version

If it does not work

The client reports an initialize or EOF error
npx can print install messages that confuse strict clients. Use -yq instead of -y in the args.
Research calls time out
Raise PERPLEXITY_TIMEOUT_MS, for example to 600000 for ten minutes.
Requests fail at work
Set PERPLEXITY_PROXY to your company proxy and check that api.perplexity.ai is allowed.
Authentication errors
Check that PERPLEXITY_API_KEY is set where the client starts the server and that the account has API credit.

Peas in a pod: similar servers

How Perplexity MCP lines up against other browser and web data servers on the shelf.

ServerToolsRuns onLicenceStars
Perplexity MCP4Your computer or hostedMIT2.6K
Tavily MCP5Your computer or hostedMIT2.5K
Playwright MCP25Your computerApache-2.038.7K
Firecrawl28Your computerMIT7.7K
Exa2Your computer or hosted–5.2K

Questions

Is this the official Perplexity server?

Yes. Perplexity publishes it as @perplexity-ai/mcp-server under the MIT licence.

Does a Perplexity Pro subscription cover it?

It uses the Perplexity API, which is billed on its own, so you need API credit.

Local or hosted?

Both have the same tools. The hosted server at api.perplexity.ai/mcp needs no install; send your key as a bearer token.

Which tool should the model use?

perplexity_search for links, perplexity_ask for quick answers, perplexity_reason for tricky questions and perplexity_research for a full report.

Full docs on GitHubperplexityai/modelcontextprotocol

Something wrong on this page? Tell us.

More in Browser and web data