perplexity-web-mcp
Provides Perplexity search using a logged-in perplexity.ai account, supporting auto, Pro, reasoning, and Deep Research modes, optional academic/scholarly search, and custom model preferences. Returns answers in Markdown with a numbered list of sources.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@perplexity-web-mcp search for the latest updates to the MCP protocol and cite sources"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
perplexity-web-mcp
An MCP server that gives AI assistants Perplexity search through your own perplexity.ai account. It drives a local Chrome with your logged-in web session, so you don't need an API key or API credits. Your Perplexity subscription limits (Pro Search, Deep Research, etc.) apply as usual.
Unofficial. This project is not affiliated with Perplexity. It uses perplexity.ai's internal web API, which can change at any time, and automated use may violate Perplexity's Terms of Service. Use it with your own account, for personal use, at your own risk.
Tool
perplexity_search(query, mode="pro", model=None, academic=False)
mode | what it does |
| fast default answer |
| Pro Search (default) |
| reasoning model |
| Deep Research: slow and thorough, counts against your daily limit |
academic=True searches scholarly sources instead of the general web. model lets you pass a raw
Perplexity model_preference value if you want a specific model.
The tool returns the answer in Markdown, followed by a numbered list of sources.
Related MCP server: Perplexity MCP
Requirements
macOS / Linux:
curl -LsSf https://astral.sh/uv/install.sh | shWindows (PowerShell):
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
Google Chrome. On Windows, Microsoft Edge is used if Chrome is not installed.
A perplexity.ai account (Pro recommended)
Install
1. Log in once
This opens a browser window. Sign in to Perplexity, then close the window.
uvx --from git+https://github.com/H1di/perplexity-web-mcp perplexity-web-mcp loginIf Google sign-in says the browser is not secure, sign in with your email instead (Perplexity sends you a code).
2. Add the server to your MCP client
Use the standard mcpServers config:
{
"mcpServers": {
"perplexity": {
"command": "uvx",
"args": ["--from", "git+https://github.com/H1di/perplexity-web-mcp", "perplexity-web-mcp"]
}
}
}Restart the client, and the perplexity_search tool becomes available.
If the client can't find uvx, use its full path as "command". This is common with desktop apps on macOS,
which don't inherit your shell PATH. To get the full path:
OS | command | typical result |
macOS |
|
|
Linux |
|
|
Windows |
|
|
In JSON on Windows, escape backslashes: "command": "C:\\Users\\you\\.local\\bin\\uvx.exe".
Platform notes
The browser runs in normal (headed) mode, because Cloudflare blocks headless browsers on perplexity.ai. It starts on the first query and closes after 60 s without queries. How it is kept out of sight depends on the OS:
OS | behavior |
Linux | With |
macOS | The window is placed off-screen. A Chrome icon appears in the Dock while queries run. |
Windows | The window is placed off-screen. It may show up in the taskbar while queries run. |
The login window is always visible.
Configuration
Set these environment variables in the server's env block:
variable | default | meaning |
| see below | browser profile that stores your login |
|
| seconds without queries before the browser closes |
|
| set to |
|
| Playwright channel to use: |
Default profile location:
Linux:
~/.local/share/perplexity-mcp/profilemacOS:
~/Library/Application Support/perplexity-mcp/profileWindows:
%LOCALAPPDATA%\perplexity-mcp\profile
If you set PPLX_PROFILE or PPLX_BROWSER, set the same value when you run login.
How it works
The server runs a real (headed) Chrome with a dedicated profile.
Queries are sent with
fetch()from inside the perplexity.ai page to the same endpoint the website uses. The server then parses the streamed response.Queries are sent in incognito mode, so they don't appear in your Perplexity history.
Troubleshooting
"No supported browser found": install Google Chrome, or set
PPLX_BROWSER."Not logged in": your session expired. Run the
logincommand again.HTTP 4xx or "Unexpected response": Perplexity probably changed its web API. Please open an issue.
Hangs on the first query: Cloudflare may be showing a challenge. Run with
PPLX_HIDE=0to watch the browser.
Development
git clone https://github.com/H1di/perplexity-web-mcp && cd perplexity-web-mcp
uv run perplexity-web-mcp login
uv run python -m unittest discover testsLicense
MIT
Available Tools
1 toolperplexity_searchA
Ask Perplexity (web search + answer with citations) using the user's Perplexity Pro subscription.
Args: query: The question or search query. mode: "auto" (fast), "pro" (Pro Search, default), "reasoning", or "deep_research" (slow, thorough). model: Optional raw Perplexity model_preference override (e.g. "pplx_pro"). academic: Search academic sources instead of the general web.
Inputs: query (required), mode, model, academic. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 1 tool
There is only one tool, so there is no possibility of misselection between competing tools. The single tool's purpose (web search with Perplexity) is unambiguous, with its modality variants folded into a `mode` argument rather than separate tools.
A single, clearly readable snake_case name (`perplexity_search`) that namespaces the provider and the action. There is no second naming convention to conflict with it.
One tool is a thin surface for a server that exposes four distinct search modes plus model overrides and academic filtering. Consolidating is reasonable, but the result sits at the low end of what an agent would expect from a dedicated search server.
For a search-and-answer domain, a single entry point that returns a cited answer covers the core workflow. Minor gaps remain: no conversation/follow-up continuation, no citation or source retrieval, and no way to poll or resume the slow `deep_research` mode.
Maintenance
Related MCP Connectors
Real-time web search, reasoning, and research through Perplexity's API
Enable AI assistants to perform web searches using Perplexity's Sonar Pro.
Live web search and fetch plus Twitter, Reddit, GitHub, Amazon, YouTube, Telegram data. No API keys.
Provides AI assistants with access to Seltz's powerful Web Search capabilities.
Related MCP Servers
- FlicenseBqualityDmaintenanceEnables AI assistants to perform real-time web and academic searches using Perplexity's Sonar API.2-
- AlicenseAqualityDmaintenanceEnables AI-powered web research using your existing Perplexity Pro subscription, providing tools for search, follow-up conversations, and thread management through the Model Context Protocol.41,201 PyPI24MIT
- AlicenseAqualityDmaintenanceEnables AI assistants to perform web searches and deep research via Perplexity, without API costs, using session-based authentication and multi-account pooling.215MIT
- AlicenseAqualityDmaintenanceProvides AI-powered research capabilities through Perplexity's web interface, enabling web search, deep research, chat, and documentation retrieval without API keys.8GPL 3.0