🔍 Perplexity Web Wrapper API & MCP

Unofficial Perplexity AI client that works without the official API. Fork of saiteja-madha/perplexity-web-wrapper, updated for the current Perplexity website.

Related: notebooklm-mcp-cli — Google NotebookLM as an MCP server

Warning

This project may not comply with Perplexity AI's Terms of Service. It works by automating a browser session against the Perplexity website — not through any official API. Use at your own risk. Your account may be rate-limited or banned. The author is not responsible for any consequences.

✨ What's New

Feature	Description
🤖 MCP Server	11 tools for Claude Code and other AI assistants (stdio)
💻 CLI	Full-featured command-line interface
📰 Discover Feed	Browse trending topics and articles
📂 Spaces	Full CRUD — create, edit, delete, add links, list threads
📝 Deep Research	Captures the full report, not just the summary
🔐 Auto-login	Playwright + stealth browser login, no manual cookie export
🍪 Cookie Persistence	Session stays fresh across restarts automatically

---

🚀 Setup

CLI

Quick install (global, no clone needed):

uv tool install git+https://github.com/eryawww/perplexity-web-wrapper-mcp.git
playwright install chromium

To update or uninstall:

uv tool install git+https://github.com/eryawww/perplexity-web-wrapper-mcp.git --force  # update
uv tool uninstall perplexity-web-wrapper                                                # remove

📦 Development install (if you want to modify the code)

git clone https://github.com/eryawww/perplexity-web-wrapper-mcp.git
cd perplexity-web-wrapper-mcp
uv sync
uv run playwright install chromium
# use `uv run perplexity` instead of `perplexity`

MCP & Claude Code

Step 1 — Install the package:

uv tool install git+https://github.com/eryawww/perplexity-web-wrapper-mcp.git
playwright install chromium

Step 2 — Register with Claude Code:

claude mcp add perplexity -- perplexity-mcp

Step 3 — Start Claude Code and try:

"Search Perplexity for the latest AI news"

On first use, a browser window opens for you to log in. After that, it's fully automatic.

To remove:

claude mcp remove perplexity
uv tool uninstall perplexity-web-wrapper

⚙️ Manual MCP config (for other clients)

{
  "mcpServers": {
    "perplexity": {
      "command": "perplexity-mcp"
    }
  }
}

---

📖 Usage

CLI

# 🔐 Login (or auto-triggered on first use)
perplexity login

# 🔎 Search
perplexity search "what is quantum computing"
perplexity search "compare React vs Vue" --mode pro
perplexity search "explain transformers" --mode "deep research"
perplexity search "follow up question" --follow-up BACKEND_UUID

# 🧵 Threads
perplexity threads
perplexity threads --search "python" --limit 5
perplexity thread SLUG

# 📰 Discover
perplexity discover

# 📂 Spaces
perplexity spaces
perplexity space SLUG
perplexity space:create "My Research" --instructions "Focus on academic sources"
perplexity space:edit UUID --title "New Title" --instructions "Be concise"
perplexity space:add-link UUID "docs.python.org"
perplexity space:threads SLUG
perplexity space:delete UUID

Claude Code Prompts

Once the MCP server is registered, you can ask Claude directly:

Prompt	What happens
"Search Perplexity for Redis vs Memcached benchmarks"	Calls search with mode auto
"Do a deep research on Python async HTTP libraries"	Calls search with mode deep research
"Show me trending topics on Perplexity"	Calls discover_feed
"List my Perplexity spaces"	Calls list_spaces
"Create a space called ML Papers with instructions to focus on arxiv"	Calls create_space
"Add docs.python.org as a source to space UUID"	Calls add_space_link

---

🛠️ APIs and Tools

Available API

The Python library can be used directly:

from lib.perplexity import Client

client = Client(cookies={}, cookies_path="perplexity_cookies.json")

Method	Description	Example
search()	Query with modes, models, sources, follow-ups	client.search("query", mode="pro", model="claude 3.7 sonnet")
get_threads()	List conversation threads	client.get_threads(limit=10, search_term="python")
get_thread_details_by_slug()	Get full thread history	client.get_thread_details_by_slug("thread-slug-abc")
get_discover_feed()	Trending topics and articles	client.get_discover_feed()
list_spaces()	All spaces grouped by category	client.list_spaces()
get_space()	Space details, instructions, links	client.get_space("my-space-slug")
create_space()	Create a new space	client.create_space("Title", instructions="Be concise")
edit_space()	Update space settings	client.edit_space(uuid, instructions="New instructions")
delete_space()	Delete a space	client.delete_space(uuid)
add_space_link()	Add focused web source	client.add_space_link(uuid, "docs.python.org")
list_space_threads()	Threads within a space	client.list_space_threads("my-space-slug")

🔎 Search Modes & Models

Mode	Available Models	Description
auto	(default)	Quick answers
pro	sonar · gpt-4.5 · gpt-4o · claude 3.7 sonnet · gemini 2.0 flash · grok-2	Enhanced search with model selection
reasoning	r1 · o3-mini · claude 3.7 sonnet	Step-by-step reasoning
deep research	(default)	Multi-step research with full report

Available Tools

MCP tools exposed via perplexity-mcp (stdio transport):

Tool	Arguments	Description
search	query, mode?, model?, sources?, language?, backend_uuid?, incognito?	Search Perplexity AI
get_threads	limit?, offset?, search_term?	List conversation threads
get_thread	slug	Get thread details
discover_feed	next_token?	Browse trending topics
list_spaces	—	List all Spaces
get_space	collection_slug	Get Space details
create_space	title, description?, instructions?, emoji?, access?	Create a Space
edit_space	collection_uuid, title?, description?, instructions?, emoji?, access?	Update Space settings
delete_space	collection_uuid	Delete a Space
add_space_link	collection_uuid, link	Add focused web link
list_space_threads	collection_slug, limit?, offset?, filter_by_user?	List threads in a Space

---

🙏 Acknowledgements

Built on top of saiteja-madha/perplexity-web-wrapper which provided the original Perplexity web client and FastAPI server.

📝 Notes

• This is an unofficial project, not affiliated with Perplexity AI.
• Requires Python 3.13+.
• Uses curl-cffi with Chrome impersonation to bypass bot detection.
• An optional FastAPI server is also included — run with uvicorn api.main:app --reload and see /docs for Swagger UI.

📄 License

MIT