Reolink Mcp

io.github.ed-dryha由 GitHub 账号 ed-dryha 发布v1.0.0更新于 Oct 24, 2026

MCP server for Reolink cameras: snapshots, device state, AI detection, PTZ, and deterrence controls

已验证STDIO仅桌面位置与生活安全与监控

概览

AI 生成的概览

让助手连接局域网内的 Reolink 摄像头,获取实时快照、设备状态、AI 侦测信息,并控制云台与威慑功能。

可以做什么
  • 列出摄像头,读取设备信息、支持能力与当前状态
  • 轮询 AI 侦测状态(人、车、宠物)及移动标记
  • 抓取实时快照并以图像返回
  • 鸣响或停止警笛,开关音频警报
  • 开关聚光灯、红外灯与白光灯,并调整变焦
  • 列出并调用云台预置位、读取云台位置、配置云台看守
适用场景
当你在局域网中有 Reolink 摄像头,希望让助手查看实时画面、了解摄像头侦测到什么,或在不打开厂商应用和 NVR 的情况下调整灯光、警笛和云台时使用。
运行要求
以 stdio 方式在本机运行,通常通过 uvx 从 PyPI 包启动。需要局域网内可访问的 Reolink 摄像头及其管理员用户名和密码。摄像头地址和用户名写在由 RMCP_CONFIG_FILE 指定的 YAML 配置文件中,每个摄像头的密码通过各自的 RMCP_CAMERAS PASSWORD 环境变量提供。
安装前请注意
服务器会为每个摄像头保持一条长连接会话,若 NVR 或厂商应用同时连接,可能触及摄像头的并发会话上限。控制类工具会改变摄像头状态,其中 set_siren 被标记为破坏性操作并会发出真实警报声。摄像头密码以环境变量传入,请勿写入共享的客户端配置。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Reolink Mcp,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

reolink-mcp

An open-source MCP (Model Context Protocol) server for Reolink cameras: lets Claude — or any MCP client — see and control Reolink cameras on the local network. Snapshots, device state, AI detection states, PTZ presets, and deterrence controls (spotlight, siren, IR/white LED). Local network only, no cloud.

Market gap (verified 2026-07-08): no dedicated Reolink MCP server exists on GitHub, npm, or PyPI. Closest prior art requires a middleman: dedsxc/mcp-frigate (needs Frigate NVR), homeassistant-mcp (needs Home Assistant). This is a first-mover standalone niche.

A user with a Reolink camera adds reolink-mcp to their MCP client config, asks "show me the front door camera," and gets a live snapshot — direct to camera, no NVR or home-automation daemon in between.

Quickstart

Requires a Reolink camera reachable on the local network and its admin username/password.

1. Create your camera config

Copy config.example.yaml to ~/.config/reolink-mcp/config.yaml (or point RMCP_CONFIG_FILE at a copy anywhere else on disk):

yaml
cameras:  front_door:    host: 192.168.1.44    username: admin

Passwords never go in this file. Set one RMCP_CAMERAS__<name>__PASSWORD environment variable per camera instead — <name> must exactly match the camera's key in the YAML above (lowercase snake_case):

bash
RMCP_CAMERAS__front_door__PASSWORD=<camera-password>

Concurrent-session limit: Reolink cameras allow only a limited number of simultaneous sessions. reolink-mcp holds one reolink-aio session per configured camera for as long as the server runs — running it alongside another system that also holds sessions on the same cameras (an NVR, another MCP server, etc.) can exhaust that limit. See Camera compatibility & session limits below.

2. Add it to your MCP client

Claude Code:

bash
claude mcp add reolink -- uvx reolink-mcp

Claude Desktop (claude_desktop_config.json — macOS: ~/Library/Application Support/Claude/claude_desktop_config.json, Windows: %APPDATA%\Claude\claude_desktop_config.json):

json
{  "mcpServers": {    "reolink": {      "command": "uvx",      "args": ["reolink-mcp"],      "env": {        "RMCP_CAMERAS__front_door__PASSWORD": "<camera-password>"      }    }  }}

Any other MCP client (generic stdio):

json
{  "command": "uvx",  "args": ["reolink-mcp"]}

That's it — ask your MCP client to "show me the front door camera."

Tools

Built on reolink-aio and the official MCP Python SDK. Every tool carries readOnlyHint/destructiveHint annotations so MCP clients can gate approvals — Observe tools are read-only, Control tools mutate camera state, and set_siren is the only tool marked destructive.

ToolTypePurposeDestructive
list_camerasObserveEnumerate every configured camera with live connection status, model, and hostNo
get_device_infoObserveModel, firmware, hardware details for a cameraNo
get_capabilitiesObserveWhat a camera supports — PTZ, siren, spotlight, IR/white LED, zoom, AI detection types — in neutral feature vocabularyNo
get_statesObserveCurrent device state: day/night mode, white LED/spotlight, IR lights, siren capability, and motionNo
get_recent_eventsObserveCurrent AI detection state (person/vehicle/pet plus any camera-reported extras) from an on-demand poll, plus the plain motion flagNo
get_snapshotObserveLive still image from a camera, returned as an MCP image content blockNo
set_sirenControlSound or stop a camera's siren — ~5s safe default, 60s hard cap, over-cap requests refused not clampedYes
set_audio_alarmControlEnable or disable a camera's siren/audio-alarm feature — needed when set_siren is accepted by the firmware but produces no soundNo
set_spotlightControlTurn a camera's spotlight on or offNo
set_ir_lightsControlSet a camera's IR lights to auto, on, or offNo
set_white_ledControlTurn a camera's white LED on or off, with optional brightnessNo
set_zoomControlZoom a camera to an absolute position (0-100) or a relative in/out stepNo
list_presetsControlList a camera's named PTZ presetsNo
ptz_move_to_presetControlMove a camera to a named (or numeric) PTZ presetNo
ptz_positionControlRead a camera's current pan/tilt/zoom position, naming the nearest saved preset when closeNo
ptz_guardControlConfigure a camera's PTZ guard point and auto-return (set/goto/enable/disable)No

Safety

  • RMCP_READ_ONLY=true starts the server with all 10 control tools stripped at startup — they are simply never registered, not hidden. A one-line notice is printed to stderr; the client sees a clean, observe-only server with no control tools in its registry at all.
  • set_siren defaults to a ~5s burst when no duration is given, and refuses (never silently clamps) any request over its 60-second hard cap — a clear error asks for a shorter duration instead.
  • Capability gating: every control tool checks the target camera's actual capabilities before sending any command, and refuses cleanly (e.g. "camera 'front_door' has no siren") instead of letting a raw API error from an unsupported feature reach the client.

Camera compatibility & session limits

Validated live during development on:

CameraValidated capabilities
P437Siren, spotlight, zoom
P320IR lights

PTZ tools (list_presets, ptz_move_to_preset, ptz_position, ptz_guard) are mock-validated only — real PTZ hardware validation is pending.

Concurrent-session limit: reolink-aio maintains one long-lived session per camera for the lifetime of the server process. Reolink cameras cap the number of simultaneous sessions a camera will accept; running reolink-mcp alongside another system that independently holds sessions on the same cameras (an NVR, Reolink's own app, a second automation stack) can hit that cap. This project's own development setup runs against cameras shared with a second, independent system and has been exercised under that exact coexistence condition — if you see login/session errors, check what else currently holds a session on the camera first.

License

MIT

来源:README.md,提交 9293a07

工具

0
SourceWeft 尚未在沙箱中启动过该软件包,因此其工具未知。

版本历史

1
  1. v1.0.0最新Sep 16, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分