perplexity-webui-mcp
MCP server that queries Perplexity Pro via WebUI session token, providing model-specific tools for search and deep research.
README
<h1 align="">perplexity-webui-mcp</h1>
<p align=""> mcp server for querying perplexity pro via webui session token. </p>
<p align=""> <img src="https://img.shields.io/badge/license-MIT-green" alt="license"> <img src="https://img.shields.io/badge/language-typescript-blue" alt="language"> <img src="https://img.shields.io/badge/npm-perplexity--webui--mcp-orange" alt="npm"> <img src="https://img.shields.io/badge/mcp-sdk-orange" alt="mcp"> <a href="https://github.com/Microck/opencode-studio"><img src="https://img.shields.io/badge/opencode-studio-brown?logo=data%3Aimage%2Fpng%3Bbase64%2CiVBORw0KGgoAAAANSUhEUgAAAA4AAAAOCAYAAAAfSC3RAAABiElEQVR4nF2Sv0tWcRTGPyeVIpCWwmyJGqQagsqCsL2hhobsD3BvdWhoj%2F6CiIKaoqXBdMjKRWwQgqZ%2BokSvkIhg9BOT9xPn9Vx79cD3cu6953zP8zznCQB1V0S01d3AKeAKcBVYA94DjyJioru2k9SHE%2Bqc%2Bkd9rL7yf7TUm%2BpQ05yPUM%2Bo626Pp%2BqE2q7GGfWrOpjNnWnAOPAGeAK8Bb4U5D3AJ%2BAQsAAMAHfVvl7gIrAf2Kjiz8BZYB3YC%2FwFpoGDwHfgEnA0oU7tgHiheEShyXxY%2FVn%2Fn6ljye8DcBiYAloRcV3tAdrV1xMRG%2Bo94DywCAwmx33AJHASWK7iiAjzNFOBl7WapPYtYdyo8RlLqVpOVPvq9KoH1NUuOneycaRefqnP1ftdUyiOt5KS%2BqLWdDpVzTXMl5It4Jr6u%2BQ%2FnhyBc8C7jpowGxGvmxuPqT9qyYuFIKdP71B8WT3SOKexXLrntvqxq3BefaiuFMQ0wqZftxl3M78MjBasfiDN%2FSAi0kFbtf8ACtKBWZBDoJEAAAAASUVORK5CYII%3D" alt="Add with OpenCode Studio" /></a> </p>
quick start
this package is a local mcp wrapper (stdio transport) that launches the upstream perplexity-webui-scraper mcp server via uvx.
by default it pins the upstream runner to:
perplexity-webui-scraper[mcp]@git+https://github.com/henrique-coder/perplexity-webui-scraper.git@prod
override that only if you need to test a different upstream build:
PERPLEXITY_UPSTREAM_FROM="perplexity-webui-scraper[mcp]@git+https://github.com/henrique-coder/perplexity-webui-scraper.git@<ref>" \
PERPLEXITY_SESSION_TOKEN="your_token_here" \
npx perplexity-webui-mcp
manual run:
PERPLEXITY_SESSION_TOKEN="your_token_here" npx perplexity-webui-mcp
manual run through a proxy:
PERPLEXITY_SESSION_TOKEN="your_token_here" \
PERPLEXITY_PROXY_URL="socks5://127.0.0.1:1080" \
npx perplexity-webui-mcp
manual run with flaresolverr:
docker run -d --name flaresolverr -p 8191:8191 ghcr.io/flaresolverr/flaresolverr:latest
PERPLEXITY_SESSION_TOKEN="your_token_here" \
PERPLEXITY_FLARESOLVERR_URL="http://127.0.0.1:8191" \
npx perplexity-webui-mcp
important: this uses perplexity's internal webui api with a session cookie. for personal/local tinkering only - not affiliated with perplexity ai.
overview
perplexity-webui-mcp is a local stdio mcp wrapper that launches the upstream perplexity-webui-scraper mcp server through uvx. this keeps the package on npm while using the upstream webui implementation for browser impersonation, retry logic, model-specific tools, and token tooling.
quick installation
paste this into your llm agent session:
Install and configure perplexity-webui-mcp by following the instructions here:
https://raw.githubusercontent.com/Microck/perplexity-webui-mcp/refs/heads/master/INSTALL.md
npm (recommended)
npm install -g perplexity-webui-mcp
runtime requirement:
uv --version
if uv is missing, install it from https://docs.astral.sh/uv/getting-started/installation/
manual installation
from source
git clone https://github.com/Microck/perplexity-webui-mcp.git
cd perplexity-webui-mcp
npm install
npm run build
getting your session token
fastest method (automatic via CLI):
uvx --with rich --from "perplexity-webui-scraper@git+https://github.com/henrique-coder/perplexity-webui-scraper.git@prod" get-perplexity-session-token
this interactive CLI asks for your email, handles OTP/magic-link verification, and prints the session token.
you can run that command from any directory.
manual method (browser):
- open perplexity.ai in your browser and log in
- open devtools (f12 or cmd+opt+i)
- go to application > cookies >
https://www.perplexity.ai - copy the value of
__Secure-next-auth.session-token
powered by token extraction flow from: https://github.com/henrique-coder/perplexity-webui-scraper
configuration
because this server uses stdio, you configure it as a local command and pass the token via env.
proxy support:
- standard proxy env vars already pass through:
HTTP_PROXY,HTTPS_PROXY,ALL_PROXY,NO_PROXYand their lowercase variants - for a simpler single-value setup, set
PERPLEXITY_PROXY_URL; the wrapper expands it into the standard proxy env vars before launching the upstream MCP server - optional bypass list:
PERPLEXITY_NO_PROXY - if perplexity keeps returning cloudflare challenge pages, set
PERPLEXITY_FLARESOLVERR_URLto a running flaresolverr instance. the wrapper solveshttps://www.perplexity.ai/search/new, injects the returned cookies into the upstreamcurl_cffisession, and stops forwarding the standard proxy env vars in that mode - optional flaresolverr overrides:
PERPLEXITY_FLARESOLVERR_SOLVE_URL- alternate url to solve first. default:https://www.perplexity.ai/search/newPERPLEXITY_FLARESOLVERR_MAX_TIMEOUT- flaresolverr solve timeout in milliseconds. default:60000
note: deep research can take longer than 60 seconds. if your client supports it, set a higher timeout (example: 10 minutes).
mcp client config (claude desktop, opencode, etc)
{
"mcpServers": {
"perplexity": {
"command": "perplexity-webui-mcp",
"timeout": 600000,
"env": {
"PERPLEXITY_SESSION_TOKEN": "your_session_token_here",
"PERPLEXITY_FLARESOLVERR_URL": "http://127.0.0.1:8191"
}
}
}
}
from source
{
"mcpServers": {
"perplexity": {
"command": "node",
"args": ["/path/to/perplexity-webui-mcp/dist/index.js"],
"timeout": 600000,
"env": {
"PERPLEXITY_SESSION_TOKEN": "your_session_token_here",
"PERPLEXITY_FLARESOLVERR_URL": "http://127.0.0.1:8191"
}
}
}
}
using flaresolverr when cloudflare blocks your host
start flaresolverr:
docker run -d --name flaresolverr -p 8191:8191 ghcr.io/flaresolverr/flaresolverr:latest
curl http://127.0.0.1:8191/
then point the wrapper at it:
PERPLEXITY_SESSION_TOKEN="your_token_here" \
PERPLEXITY_FLARESOLVERR_URL="http://127.0.0.1:8191" \
npx perplexity-webui-mcp
flaresolverr mode:
- the wrapper solves the cloudflare wall in flaresolverr first, then injects the returned cookies into the upstream session
- the wrapper does not forward
HTTP_PROXY,HTTPS_PROXY,ALL_PROXY, orNO_PROXYto the upstream process in this mode because the solved cookies need to stay tied to the flaresolverr browser route - if flaresolverr itself needs a proxy, configure that on the flaresolverr side instead of on
perplexity-webui-mcp
remote deployment over tailscale (optional)
if your cloud machine gets blocked by cloudflare but your home machine works, run the upstream mcp server on the home machine and connect to it from opencode as a remote mcp.
- copy templates from this repo:
deploy/systemd/perplexity-webui-mcp.env.exampledeploy/systemd/perplexity-webui-mcp-sse.shdeploy/systemd/perplexity-webui-mcp.service
- install and enable service on the home machine (user service):
mkdir -p ~/.config ~/.config/systemd/user ~/.local/bin
cp deploy/systemd/perplexity-webui-mcp.env.example ~/.config/perplexity-webui-mcp.env
cp deploy/systemd/perplexity-webui-mcp-sse.sh ~/.local/bin/perplexity-webui-mcp-sse.sh
cp deploy/systemd/perplexity-webui-mcp.service ~/.config/systemd/user/perplexity-webui-mcp.service
chmod 600 ~/.config/perplexity-webui-mcp.env
chmod 755 ~/.local/bin/perplexity-webui-mcp-sse.sh
systemctl --user daemon-reload
systemctl --user enable --now perplexity-webui-mcp.service
- point opencode (cloud host) to the tailscale endpoint:
{
"mcp": {
"perplexity-webui": {
"type": "remote",
"url": "http://<tailscale-ip>:8790/sse",
"enabled": true,
"timeout": 600000,
"oauth": false
}
}
}
- verify:
opencode mcp list
features
| tool | description |
|---|---|
pplx_ask |
best-model query (auto model selection) |
pplx_deep_research |
deep research mode |
pplx_sonar |
sonar model |
pplx_gpt54 / pplx_gpt54_thinking |
gpt-5.4 variants |
pplx_claude_o46 / pplx_claude_o46_think |
claude opus 4.6 variants |
pplx_claude_s46 / pplx_claude_s46_think |
claude sonnet 4.6 variants |
pplx_gemini31_pro / pplx_gemini31_pro_think |
gemini 3.1 pro variants |
pplx_gemini_flash / pplx_gemini_flash_think |
gemini flash variants |
pplx_grok41 / pplx_grok41_think |
grok 4.1 variants |
pplx_nemotron3_super_think |
nemotron 3 super thinking |
all upstream model tools support source_focus values: web, academic, social, finance, all.
how this differs from v1.0.0
- old v1.0.0: one custom tool (
perplexity_search) implemented in local TypeScript HTTP logic. - current: delegates to upstream
perplexity-webui-scraperMCP, exposing the full upstream model-specific toolset. - result: significantly better compatibility with Perplexity anti-bot protections.
troubleshooting
| problem | solution |
|---|---|
| token invalid / 401 | get a fresh token from browser cookies |
uvx not found |
install uv (uv --version should work) |
| no answer returned | check rate limits or whether your account can access the selected model |
| clarifying questions error | deep research retries once with recommended/default clarification choices; if this still appears, Perplexity rejected the skip and needs a more specific query |
cloudflare challenge / Just a moment... |
run flaresolverr and set PERPLEXITY_FLARESOLVERR_URL=http://127.0.0.1:8191 |
| timeout | deep research can take several minutes - be patient |
verify both modes quickly
PERPLEXITY_SESSION_TOKEN="your_token_here" npm run self-test
this checks both:
- regular search (
best) - deep research (
deep-research)
and prints pass/fail per mode.
project structure
perplexity-webui-mcp/
├── deploy/
│ └── systemd/
│ ├── perplexity-webui-mcp.env.example
│ ├── perplexity-webui-mcp-sse.sh
│ └── perplexity-webui-mcp.service
├── src/
│ └── index.ts # proxy launcher for upstream MCP
├── package.json
├── tsconfig.json
├── .env.example
├── .gitignore
├── LICENSE
├── INSTALL.md
└── README.md
license
mit
author
shoutout
special thanks to henrique-coder/perplexity-webui-scraper for the WebUI reverse-engineering and token CLI workflow that helped this project.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。