关于 Jellyfin MCP Server
Jellyfin MCP Server 是面向 AI 助手的 Jellyfin 插件,可搜索媒体、控制播放并管理媒体库内容。每次调用均受令牌范围和 Jellyfin 用户权限限制,管理员工具也单独授权。
主要功能
- 01搜索和浏览媒体库
- 02控制播放并查看会话
- 03管理播放列表和收藏
- 04运行管理员服务器任务
- 05按范围和用户权限访问
适用场景
- 01在 Jellyfin 媒体库中查找影视内容
- 02控制设备上的播放
- 03更新个人收藏或播放列表
- 04以管理员身份运行媒体库和元数据任务
MCP 相关
插件通过无状态 Streamable HTTP 在 `POST /mcp` 向 MCP 客户端提供 Jellyfin 工具。令牌决定可用范围,每次调用使用关联 Jellyfin 用户的权限。
配置
远程服务无需本地安装。选择你使用的客户端,复制配置或使用一键安装。
https://{jellyfin_host}/mcpclaude mcp add --transport http aelui-jellyfin-plugin https://{jellyfin_host}/mcp{"mcpServers":{"aelui-jellyfin-plugin":{"type":"http","url":"https://{jellyfin_host}/mcp"}}}在 Cursor 中打开{"servers":{"aelui-jellyfin-plugin":{"type":"http","url":"https://{jellyfin_host}/mcp"}}}在 VS Code 中打开{"mcpServers":{"aelui-jellyfin-plugin":{"type":"http","url":"https://{jellyfin_host}/mcp"}}}{"mcpServers":{"aelui-jellyfin-plugin":{"serverUrl":"https://{jellyfin_host}/mcp"}}}命令在终端中运行;JSON 粘贴到对应客户端的 MCP 配置后重启。需要授权的服务会在首次连接时提示登录。
https://{jellyfin_host}/mcp需要授权工具
README 原文English
来自项目 READMEREADME 暂无中文译文,以下为英文原文;中文介绍见上方。
Jellyfin MCP Server plugin

A Jellyfin plugin that exposes your server to AI assistants through the Model Context Protocol (MCP). Assistants such as Claude can search and browse your library, see what you are watching, control playback on your devices, manage playlists and favorites, and (for administrators) run server tasks. Every call runs as the Jellyfin user who owns the access token, limited to that user's permissions and to the scopes granted to the token. The plugin has no runtime dependencies besides Jellyfin itself.
Compatibility
- Jellyfin 12.x only (compiled against 12.4.0,
targetAbi12.4.0.0). Older servers are not supported. - Built with .NET 10.
- MCP transport: Streamable HTTP, stateless,
POST /mcp. Protocol revisions 2025-11-25, 2025-06-18 and 2025-03-26. The stateless 2026-07-28 revision is not supported yet (see the roadmap). - Clients that send a static
Authorization: Bearerheader work; Claude Code is the supported client. Claude Desktop throughmcp-remoteis expected to work but is not yet verified. claude.ai web connectors need OAuth and are not supported yet.
Install in three steps
- In Jellyfin go to Dashboard > Plugins > Repositories > + and add
https://raw.githubusercontent.com/AeLui/jellyfin-plugin-mcp/main/manifest.json. - Open the Catalog, install MCP Server and restart Jellyfin.
- Open Dashboard > Plugins > MCP Server, create a token (pick the user and the scopes) and copy it. It is shown only once.
More detail, including a manual install, is in docs/installation.md.
Connect a client
Replace https://jellyfin.example.com with your server address and jmcp_... with your token.
Claude Code, with the plugin from this repository's marketplace (connects the server and adds the agent skill; it asks for the URL and the token, and keeps the token in your system's credential store):
/plugin marketplace add AeLui/jellyfin-plugin-mcp
/plugin install jellyfin-mcp@jellyfin-plugin-mcp
On Claude Code 2.1.275 or later, one step is enough: /plugin install jellyfin-mcp --marketplace AeLui/jellyfin-plugin-mcp. To change the URL or token later, run /plugin configure jellyfin-mcp@jellyfin-plugin-mcp.
Claude Code, server only:
claude mcp add --transport http jellyfin https://jellyfin.example.com/mcp \
--header "Authorization: Bearer jmcp_..."
Claude Desktop (claude_desktop_config.json; requires Node.js because it runs npx mcp-remote):
{
"mcpServers": {
"jellyfin": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://jellyfin.example.com/mcp",
"--header", "Authorization: Bearer jmcp_..."]
}
}
}
The token dialog in the plugin page generates both snippets for you. The Claude Desktop flow is expected to work but has not been verified against a live server yet.
Scopes and tools
22 tools in four scopes. A token only sees the tools its scopes allow.
| Scope | Tools | Purpose |
|---|---|---|
library | 9 | Read-only: search, browse, details, people, recent, next up, continue watching, history |
playback | 4 | List sessions, play, control playback, show a message |
manage | 4 | The user's own watch state and favorites, plus playlists the user owns or may edit |
admin | 5 | Users, scheduled tasks, library scan, metadata refresh (administrators only) |
Full reference: docs/tools.md.
Agent skill
skills/jellyfin-mcp/SKILL.md is an optional Agent Skill that teaches an assistant when and how to use these tools: getting ids before acting, picking a playback session, paging, and asking before destructive actions.
The Claude Code plugin above includes it. Other agents: copy the skills/jellyfin-mcp folder into their skills directory (for Claude Code without the plugin, ~/.claude/skills/ or .claude/skills/ in a project). If you install the plugin, remove any server you added with claude mcp add, or the tools show up twice.
Privacy
The plugin runs inside your Jellyfin server and makes no outbound connections. The Claude Code plugin connects only to the Jellyfin URL you configure. What your assistant does with the tool results is governed by the assistant's own privacy policy.
Documentation
- Installation and client setup
- Tool reference
- Security model - read this before exposing the endpoint beyond your LAN
- Architecture
- Development
- Roadmap
Support the project
The plugin is free and open source. If it is useful to you, you can support its development through GitHub Sponsors.
License
GPL-3.0, see LICENSE.
评论
一个开放协议,规定了 AI 应用如何以统一方式连接外部工具与数据源。客户端负责与模型对话,MCP 服务负责提供工具、资源和提示模板,两者通过标准消息通信,同一个服务可以被不同客户端复用。