lutris-source-mcp
Closes the find-download-install loop for lutris-mcp by searching Prowlarr, downloading via qBittorrent, polling until complete, classifying the result, and returning a path for installation.
README
lutris-source-mcp
Source pipeline MCP for lutris-mcp.
Closes the find -> download -> install loop: searches Prowlarr, hands off
to qBittorrent, polls until done, classifies the resulting tree, returns a
path that lutris-mcp's install_from_yaml or install_from_directory
consumes directly.
Tools
| Tool | What it does |
|---|---|
prepare_install_source(query, ...) |
The composite. Search → grab → poll → classify → return path. mutates: true |
cancel_pipeline(infohash) |
Abort an in-flight grab. mutates: true |
stop_seeding(infohash) |
Remove from qBittorrent without deleting on-disk files. mutates: true |
pipeline_health() |
Verify Prowlarr + qBittorrent reachable; surface configured indexers. mutates: false |
setup(...) |
Write ~/.config/lutris-source-mcp/config.toml (first-run). mutates: true |
Multi-tracker ranking
Prowlarr aggregates public (1337x, TPB, etc.) and private (Gazelle: RED, OPS, GGn) trackers in one search. The composite ranks across them with this key:
(indexer_priority, # 1 = highest, 50 = lowest
freeleech_first if freeleech_only else 0,
-seeders,
size_bytes) # smaller wins as tiebreak
indexer_priority comes from the Prowlarr indexer's priority field — set
it in Prowlarr's UI per tracker (e.g. private = 1, public = 25). The MCP
respects it rather than inventing a separate trust ranking.
Gazelle hygiene
Pass freeleech_only=True (or set policy.freeleech_only = true in TOML)
to drop non-freeleech results from Gazelle indexers before ranking. Public
results pass through unchanged (freeleech is meaningless there). Auth
itself stays in Prowlarr's UI; the MCP never sees Gazelle credentials.
Stall detection
Each poll prints a stdout heartbeat:
[prepare_install_source] deadbeef dl=12345/100000 (12.3%) peers=8 rate=215.4KiB/s elapsed=14s
If downloaded doesn't increase for stall_timeout_seconds (default 5min),
the tool returns {ok: false, reason: "stalled"} so the agent can pick
another candidate.
Config
~/.config/lutris-source-mcp/config.toml:
[prowlarr]
base_url = "http://localhost:9696"
api_key = "..."
[qbittorrent]
base_url = "http://localhost:8080"
username = "admin"
password = "..."
download_dir = "/mnt/games/Downloads"
[policy]
blocklist = ["fitgirl"]
allow_indexers = []
max_size_gb = 50
freeleech_only = false
min_seeders = 1
Every key has a LUTRIS_SOURCE_MCP_<SECTION>_<KEY> env override
(e.g. LUTRIS_SOURCE_MCP_PROWLARR_API_KEY, LUTRIS_SOURCE_MCP_POLICY_FREELEECH_ONLY).
Companions
For read-only Prowlarr/qBittorrent inspection, install
jmagar/prowlarr-mcp and
pickpppcc/qbittorrent-mcp
alongside. Three small focused servers, no overlap.
License
MIT.
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。