Torrent Search MCP/API/WebUI

This repository provides a Python API/WebUI and an MCP (Model Context Protocol) server to find torrents programmatically on ThePirateBay, 1337x, Nyaa, YTS, EZTV, FitGirl, SubsPlease and UIndex. It allows for easy integration into other applications or services.

<div align="center" style="margin: 20px 0;"> <img src=".github/assets/cover.png" alt="Torrent Search web UI - popular torrents view" width="720" /> </div>

Quickstart

How to use it with MCP Clients

Run it with Docker to bypass common DNS issues

Search directly from the command line

uvx torrent-search-mcp --mode cli "sample show"

# MCP server over stdio (default)
uvx torrent-search-mcp --mode stdio

# MCP server over streamable HTTP (port 8000, endpoint /mcp)
uvx torrent-search-mcp --mode http

# MCP server over SSE (port 8000, endpoint /sse, legacy)
uvx torrent-search-mcp --mode sse

# Standalone API server (port 8000)
uvx torrent-search-mcp --mode api

Table of Contents

• Features
• Supported Sources
• Setup
• Prerequisites
• Configuration
• Installation
• Install from PyPI (Recommended)
• For Local Development
• For Docker
• Usage
• As CLI
• As Python Wrapper
• As MCP Server
• As API Server
• Via MCP Clients
• Example with Devin
• Changelog
• Contributing
• License

Features

• API wrapper for ThePirateBay, 1337x, Nyaa, YTS, EZTV, FitGirl, SubsPlease and UIndex.
• MCP server interface (FastMCP 4) serving the 2026-07-28 protocol revision over stdio or streamable HTTP (http), with automatic negotiation of older handshake revisions and legacy transport aliases (streamable-http, sse).
• API server interface for alternative HTTP access (e.g., for direct API calls or testing).
• CLI mode for quick one-off searches directly from the terminal.
• popular_torrents cached for 2 minutes for Web UI delivery (a Web UI POPULAR click passes refresh=true to drop the caches and re-scrape every source); searches are never cached (only identical concurrent requests are coalesced) so new results appear immediately.
• In-memory 1-hour, 5000-entry torrent cache used only to resolve magnet links via get_torrent without re-scraping.
• Magnet links are always stored internally; MCP hides them unless INCLUDE_LINKS=true.
• Configurable source filtering via environment variables.
• Telegram-gated web UI: one-time QR/deep-link pairing, forward-to-Telegram popup and (optional) server-side forwarding.
• Tools:
• Search for torrents across all available sources.
• Get the most popular torrents per source (apibay, uindex, 1337x, YTS, nyaa, EZTV).
• Get the magnet link for a specific torrent by id.
• List available sources.
• Present the web UI and its Telegram pairing access, approve pairing codes, and forward torrents to Telegram chats.

Supported Sources

The API exposes public display domains where applicable: apibay.org is shown as thepiratebay.org, and yts.mx as yts.vg. Results may include a validated HTTP(S) page_url linking back to their source page.

Note on UIndex: the site exposes no programmatic search endpoint (its search path is protected by a browser challenge), so queries are matched client-side against its live top list - which conveniently carries magnet links inline.

Sources can be excluded individually via the EXCLUDE_SOURCES env var.

Setup

Prerequisites

• Python 3.10+ (required for PyPI install). CI and Docker images use Python 3.14.
• uv (for local development).
• Docker and Docker Compose (for Docker setup).

Configuration (Optional)

The application reads configuration from environment variables. The recommended way to set them is by creating a .env file in your project's root directory. The application will load it automatically. See .env.example for all available options.

Installation

Choose one of the following installation methods.

#### Install from PyPI (Recommended)

This method is best for using the package as a library or running the server without modifying the code.

1. Install the package from PyPI:

pip install torrent-search-mcp

1. Create a .env file in the directory where you'll run the application (optional).

1. Run the MCP server (default: stdio):

python -m torrent_search

#### For Local Development

This method is for contributors who want to modify the source code. Using uv:

1. Clone the repository:

git clone https://github.com/philogicae/torrent-search-mcp.git
cd torrent-search-mcp

1. Install dependencies using uv:

uv sync --frozen

1. Create your configuration file by copying the example:

cp .env.example .env

1. Run the MCP server (default: stdio):

uv run -m torrent_search

The repo also ships a dev.sh helper that locks/syncs deps, formats, lints, type-checks (ty) and runs the test suite with coverage:

./dev.sh

#### For Docker

This method uses Docker Compose to run two containers: the REST API + web UI, and an MCP server that proxies the API (no local scraping).

compose.yaml is configured to bypass DNS issues (using quad9 DNS).

1. Clone the repository (if you haven't already):

git clone https://github.com/philogicae/torrent-search-mcp.git
cd torrent-search-mcp

1. Create your configuration file by copying the example:

cp .env.example .env

1. Build and run the containers using Docker Compose:

docker compose up --build -d

1. Access container logs:

docker logs torrent-search-api -f
docker logs torrent-search-mcp -f

Usage

The package exposes a single entry point, torrent-search-mcp (installed by pip/uvx), equivalent to python -m torrent_search. It supports the following --mode values:

The server is built on FastMCP 4 (MCP SDK v2). Clients supporting the 2026-07-28 revision negotiate it automatically (stateless requests, server/discover, no session IDs); older clients fall back to the previous handshake era against the same deployment.

MCP modes (stdio, http, streamable-http, sse) run standalone by default (tools scrape locally). Set TORRENT_SEARCH_API_URL to switch to API mode: the tools proxy a running Torrent Search REST API instead.

Common flags (for http, streamable-http, sse and api modes): --host (default 0.0.0.0), --port (default 8000), --reload, --workers (API only).

As CLI

Run a one-off search directly from the terminal. Prints each result as id (seeders|leechers|downloads) - filename, then fetches the magnet/torrent for the top hit.

# Using the installed entry point
torrent-search-mcp --mode cli "sample show"

# Or via uvx without installing
uvx torrent-search-mcp --mode cli "sample show"

# Or from source
uv run -m torrent_search --mode cli "sample show"

As Python Wrapper

from torrent_search import torrent_search_api

results = await torrent_search_api.search_torrents("sample show")
for torrent in results:
    print(
        f"{torrent.filename} | {torrent.size} | {torrent.seeders} SE | {torrent.leechers} LE | {torrent.date} | {torrent.source}"
    )

search_torrents is async and accepts an optional max_items (default 20). popular_torrents(per_source=20) returns the current most popular torrents from sources with a top listing - up to per_source results per source (pass per_source=None for everything), merged and ranked by seeders + leechers. Each Torrent exposes id, filename, category, size, seeders, leechers, downloads, date, source, uploader, and, when available, magnet_link and a validated HTTP(S) page_url. Pass a torrent's id to get_torrent() to retrieve its magnet link.

As MCP Server

from torrent_search import torrent_search_mcp

torrent_search_mcp.run(transport="http")

As API Server

This project also includes a API server as an alternative way to interact with the library via a standard HTTP API. This can be useful for direct API calls, integration with other web services, or for testing purposes.

Running the API Server:

# With Python
python -m torrent_search --mode api
# With uv
uv run -m torrent_search --mode api

• --host <host>: Default: 0.0.0.0.
• --port <port>: Default: 8000.
• --reload: Enables auto-reloading when code changes (useful for development).
• --workers <workers>: Default: 1. Caching and request coalescing are in-process, so each worker keeps its own cache; use the default single worker unless per-worker caches are acceptable (a shared backend such as Redis would be needed to scale out).

The API server will then be accessible at http://<host>:<port>.

Available Endpoints: The API server exposes similar functionalities to the MCP server. Key endpoints include:

• GET /: Built-in web UI (dark/light) - search, per-site popular tiles, sortable results, source-page links and magnet actions. Telegram sending requires one-time QR pairing when configured.
• POST /torrent/search: Search for torrents. Query params: query (required), max_items (optional, default 20; uncapped when per_source is set) and per_source (optional, keep up to N results per source, ranked by swarm health).
• GET /sources: List the available torrent source domains.
• GET /torrent/popular: Get the most popular torrents. Query params: per_source (optional, default 20) and refresh (optional, default false; true drops the server-side caches and fetches a fresh listing from every source - the Web UI POPULAR button does this on each click).
• GET /torrent/{torrent_id}: Get the magnet link for a specific torrent by id. Returns the magnet URI as a JSON string.
• GET /telegram/session: Web UI auth state (enabled, authenticated, public bot handle, prune_magnet_links). Send the session token as Authorization: Bearer.
• POST /telegram/auth/challenge: Create a one-time 16-character alphanumeric pairing code (rate-limited). Codes expire after ~5 minutes and are shown as a QR + deep link in the pairing popup.
• GET /telegram/auth/poll?code=: Poll a pairing code; on approval returns the one-time session token for the browser to store.
• DELETE /telegram/auth/challenge/{code}: Cancel a pending pairing code.
• POST /telegram/auth/register: Approve a pairing code bound to a Telegram chat_id. Requires Authorization: Bearer $TORRENT_SEARCH_API_KEY.
• POST /telegram/auth/logout: Revoke the presented session token.
• POST /forward_telegram: Send torrent info to the Telegram chat bound to the presented session token; server-to-server callers may instead use Authorization: Bearer $TORRENT_SEARCH_API_KEY with the target chat_id query param. JSON body: filename (required), magnet_link (required), optional size, seeders. When PRUNE_MAGNET_LINKS=true the forwarded magnet is pruned; requires TELEGRAM_BOT_TOKEN, otherwise 503.
• /docs: Interactive API documentation (Swagger UI).
• /redoc: Alternative API documentation (ReDoc).

Environment variables are configured the same way as for the MCP server (via an .env file in the project root).

Via MCP Clients

Usable with any MCP-compatible client. Available tools:

• search_torrents(user_intent, query): Search for torrents across all available sources.
• user_intent: A short description reflecting the user's overall intention (e.g. "latest episode of Sample Show").
• query: Optimized, lowercase, space-separated keywords (e.g. "sample show s01e05"). Generic/filler/technical terms should be stripped per the tool's docstring.
• By default magnet links are stripped from the response to save tokens; set INCLUDE_LINKS=true to include them.
• Magnet round trips: with INCLUDE_LINKS unset, results keep their id but no magnet. After picking the 2-5 torrents worth recommending, call get_torrent(id) once per pick: the server resolves it from its 1-hour torrent cache, or re-runs the search if the entry expired. Do not fetch magnets for every result.
• popular_torrents(per_source=20): Get the most popular torrents right now from sources with an official top listing (apibay, uindex, 1337x, YTS, nyaa, EZTV) - up to per_source results each, grouped per source and pre-ranked by seeders + leechers.
• By default magnet links are stripped from the response to save tokens; set INCLUDE_LINKS=true to include them.
• available_sources(): Get the list of available torrent sources.
• get_torrent(torrent_id): Get the magnet link for a specific torrent by id (the id returned by search_torrents or popular_torrents).
• authorize_webapp(code, chat_id): Approve a Web UI pairing code bound to your Telegram chat id (the code shown in the browser pairing gate). Requires TORRENT_SEARCH_API_KEY and TORRENT_SEARCH_API_URL.
• forward_torrent(filename, magnet_link, chat_id, size=None, seeders=None): Forward a torrent (filename + magnet) to your Telegram chat through the REST API (magnet pruned when PRUNE_MAGNET_LINKS=true). Requires TORRENT_SEARCH_API_KEY and TORRENT_SEARCH_API_URL; rate-limited per chat (20/min).
• torrent_webapp(): Present the web UI URL (WEBUI_URL) and its pairing-based access system.

#### Example with Devin

Configuration:

{
  "mcpServers": {
    ...
    # with stdio (only requires uv)
    "torrent-search-mcp": {
      "command": "uvx",
      "args": [ "torrent-search-mcp" ]
    },
    # with streamable-http transport (Docker compose: MCP on port 8001; standalone server: 8000)
    "torrent-search-mcp": {
      "serverUrl": "http://127.0.0.1:8001/mcp"
    },
    # with sse transport (legacy; requires running server)
    "torrent-search-mcp": {
      "serverUrl": "http://127.0.0.1:8001/sse"
    },
    ...
  }
}

Changelog

See CHANGELOG.md for a history of changes to this project.

Contributing

Contributions are welcome! Please open an issue or submit a pull request.

License

This project is licensed under the MIT License - see the LICENSE file for details.