Files
MaaAssistantArknights/docs/en-us/protocol/remote-control-schema.md
Lucien Shaw 14194d1118 chore: 完善容器配置及依赖安装 (#14208)
格式化工具部分:
1. pre-commit 引入 python 的格式化工具,包括 black(代码格式化)和 isort(对“包导入顺序”的规范)
2. 允许 prettier 对文档站的 markdown 文件格式化
3. 不允许 prettier 对 markdown 文件中的代码块的代码本身进行格式化
4. 升级了 pre-commit 的各个 hook 的版本
5. 优化了 pre-commit 的日志文本显示

容器总览部分:
1. 由原来的单一轻量环境转为区分空环境、轻量环境和全量环境
2. 空环境是裸 Linux 镜像(Ubuntu),为默认环境
3. 轻量环境适合开发文档站前端
4. 全量环境适合开发 MaaCore
5. 目前,全量环境完整包含了轻量环境,轻量环境完整包含了空环境
6. 在仓库 README.md 中更新了三个环境的描述,并将链接分别设置为对应环境的创建链接
   **注意:没有修改文档站中的对应文件**
7. 在各个语言的开发指南的最后,移除了 Codespace 部分的“安装额外依赖”相关描述,且将链接设置为全量环境的创建链接
   **注意:没有添加“开发文档站”的指南和对应 codespace 的使用方式**

容器的轻量环境和全量环境共有部分:
1. 安装 black 和 isort 包
2. 调整 VS Code 设置,取消先前(对 markdown 文件单独指定 markdownlint 扩展作为格式化工具)的错误设置,现在 markdown 文件仍然使用默认的 prettier 扩展作为格式化工具
3. 引入 markdown-all-in-one 扩展作为语法提示工具
4. 将 node_modules 和 3rdparty 排除在 VS Code 的文本的搜索路径之外

容器的全量环境部分:
1. 为 tools 下的所有 python 脚本安装依赖
2. 使用 tools/maadeps-download.py 下载 maadeps,且将必要二进制文件软链接到 /usr/local/bin/
3. 使用 apt 安装 cmake 和 clangd-20,将后者通过 update-alternatives 设置为系统 clangd 的默认版本
4. 使用 cmake tools 扩展,并按照 Linux 编译方法进行配置
5. 使用 clang-format 作为 c/cpp 的格式化工具,clang-format 程序主体来自 maadeps(已经软链接到 /usr/local/bin/)
6. 使用 clangd 作为 c/cpp 的语法提示工具
7. 将 MaaDeps、install 和 build 排除在 VS Code 的文本搜索路径之外

其它手动调整:
1. 更新文档站的 package.json,指定 pnpm 包管理器的版本
2. 手动保证 markdown 文件中的列表前后有空行(注意到 MarkdownLint 官方规则不一定能精准定位所有“列表前后空行”的问题,详见:https://github.com/DavidAnson/markdownlint/blob/main/doc/Rules.md#md032---lists-should-be-surrounded-by-blank-lines )
3. 修改了部分 markdown 文件中的 json 代码块的语法问题
   **注意:相同的问题并未全部发现,仅修改了两处**
4. 在 tools 目录中,一处 python 脚本的包名误用(本地包名和某个 pip 包重名),这里修改了相应代码
5. 在 tools 目录中,一处 python 脚本使用了弃用的包 cchardet 的问题,这里更换了推荐使用的功能相近的包并修改了相应代码

自动化脚本提交的修改:
1. 自动格式化了大量 tools 中的 python 脚本
2. 自动格式化了大量 docs 中的 markdown 文件

Commits:

* chore: pre-commit引入black和isort规范py文件

* chore: Auto update by pre-commit hooks [skip changelog]

* chore: devcontainer添加isort扩展,排序python导入

* chore: pre-commit任务命名及更名

* style: isort fix

* chore: Auto update by pre-commit hooks [skip changelog]

* chore: 更新pre-commit的hook版本

* fix: 模块名与第三方库重名,大忌

* chore: 容器构建时额外安装isort

* docs: md -> markdown

* chore: 容器安装python包和maadeps

* fix: 修复过时python包

* chore: 指定pnpm版本

* chore: container支持选择轻量环境

* chore: 去掉rust

* chore: add plain env

* chore: 使用clangd语言服务器

* chore: 无需单独设置markdown的格式化工具

* chore: 更新安装的clangd版本

* docs: 简易文档适配

* docs: 在仓库README中重新编排codespaces相关指引

* chore: Auto update by pre-commit hooks [skip changelog]

* style: 调整缩进

* chore: 格式化工具不用特意排除被gitignore忽略的文件

* chore: sh文件在gitattributes中单列一类

* chore: 格式化docs下的markdown文件

* chore: 暂时不修改md文件中的代码块

* style: 人为明确markdown中的部分列表相关格式

* docs: 补上部分markdown的json代码块中缺失的逗号

* chore: Auto update by pre-commit hooks [skip changelog]

* chore: Auto update by pre-commit hooks [skip changelog]

* fix: 补上tools的服务器排序相关代码中缺失的逗号

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* chore: 使用maadeps的clangd

* build: 更新maadeps工具链版本

* style: prettier fix

* revert: 还原maadeps版本

* revert: 取消使用maadeps的clangd依赖,改用系统apt安装

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
2025-09-30 19:39:48 +08:00

10 KiB

order, icon
order icon
8 mdi:remote-desktop

Remote Control Schema

To implement remote control of MAA, you need to provide a service that must be an HTTP(S) service and offer the following two anonymously accessible endpoints. These endpoints must be HTTP(S) web endpoints.

::: warning If the endpoint uses HTTP protocol, MAA will issue a security warning with each connection. Deploying plaintext transmission services on public internet is highly discouraged and dangerous, for testing purposes only. :::

::: tip Please note that JSON files do not support comments. The comments in this document are for demonstration purposes only. Do not copy them directly. :::

Task Retrieval Endpoint

MAA continuously polls this endpoint at 1-second intervals, attempting to retrieve tasks to execute and executing them in the order received.

The endpoint path is arbitrary but must be an HTTP(S) endpoint. For example: https://your-control-host.net/maa/getTask

The controlled MAA must enter this endpoint in the Task Retrieval Endpoint text box in MAA's configuration.

This endpoint must accept a POST request with Content-Type=application/json and must accept the following JSON as request content:

{
    "user":"ea6c39eb-a45f-4d82-9ecc-33a7bf2ae4dc",          // User identifier entered in MAA settings
    "device":"f7cd9682-3de9-4eef-9137-ec124ea9e9ec"         // Device identifier automatically generated by MAA
    ...     // If your endpoint has other uses, you can add optional parameters, but MAA only transmits user and device
}

This endpoint must return a JSON-formatted response that meets at least the following format:

{
    "tasks":                            // List of tasks for MAA to execute, currently supported types shown in example, if tasks doesn't exist connection is considered invalid
    [
        // Sequential execution tasks: these tasks are queued for execution in the order issued
        {
            "id": "b353c469-b902-4357-bd8f-d133199eea31",   // Unique task ID, string type, used when reporting task status
            "type": "CaptureImage",                         // Screenshot task, captures current emulator screen and includes as Base64 string in task report payload. If issuing this task type, note your endpoint's maximum request size limit, as screenshots can be tens of MB, exceeding typical gateway defaults
        },
        {
            "id": "15be4725-5bd3-443d-8ae3-0a5ae789254c",   // Unique task ID, string type, used when reporting task status
            "type": "LinkStart",                            // Start one-click farming
        },
        {
            "id": "15be4725-5bd3-443d-8ae3-0a5ae789254c",   // Unique task ID, string type, used when reporting task status
            "type": "LinkStart-Recruiting",                 // Immediately execute specific one-click farming sub-function based on current configuration, ignoring main interface checkbox. Optional values detailed below
        },
        {
            "id": "b353c469-b902-4357-bd8f-d133199eea31",   // Unique task ID, string type, used when reporting task status
            "type": "Toolbox-GachaOnce",                    // Toolbox gacha task, optional values: Toolbox-GachaOnce, Toolbox-GachaTenTimes
        },
        {
            "id": "b353c469-b902-4357-bd8f-d133199eea31",   // Unique task ID, string type, used when reporting task status
            "type": "Settings-ConnectionAddress",           // Configuration modification task, equivalent to ConfigurationHelper.SetValue("ConnectionAddress", params); For security, not all settings can be modified, see below for allowed settings
            "params": "value"                               // Value to modify
        },
        // Immediate execution tasks: these tasks can execute during sequential tasks and MAA guarantees any of these will return results quickly, typically used for controlling the remote control function itself
        {
            "id": "b353c469-b902-4357-bd8f-d133199eea31",   // Unique task ID, string type, used when reporting task status
            "type": "CaptureImageNow",                      // Immediate screenshot task, similar to screenshot task above but executes immediately without waiting for other tasks
        },
        {
            "id": "b353c469-b902-4357-bd8f-d133199eea31",   // Unique task ID, string type, used when reporting task status
            "type": "StopTask",                             // "Stop current task" task, attempts to end currently running task. If task list contains other tasks, continues with next task. This task doesn't wait to confirm current task has stopped before returning, use heartbeat task to confirm stop command effectiveness
        },
        {
            "id": "b353c469-b902-4357-bd8f-d133199eea31",   // Unique task ID, string type, used when reporting task status
            "type": "HeartBeat",                            // Heartbeat task, returns immediately with currently executing sequential task's ID as payload, or empty string if no task executing
        }
    ],
    ...     // If your endpoint has other uses, you can add optional return values, but MAA only reads tasks
}

These tasks execute in sequence, meaning if you first send a recruitment task followed by a screenshot task, the screenshot will execute after the recruitment task completes. The endpoint should be reentrant and repeatedly return tasks to execute, as MAA automatically tracks task IDs and won't re-execute tasks with the same ID.

::: note

  • LinkStart-[TaskName] task types include LinkStart-Base, LinkStart-WakeUp, LinkStart-Combat, LinkStart-Recruiting, LinkStart-Mall, LinkStart-Mission, LinkStart-AutoRoguelike, LinkStart-Reclamation
  • Settings-[SettingsName] task types include Settings-ConnectionAddress, Settings-Stage1
  • Settings series tasks still execute sequentially rather than immediately, queuing behind previous tasks
  • Multiple immediate execution tasks also execute in issued order, though since these tasks execute quickly, their order generally doesn't matter

:::

Task Reporting Endpoint

When MAA completes a task, it reports the execution result to the remote server through this endpoint.

The endpoint path is arbitrary but must be an HTTP(S) endpoint. For example: https://your-control-host.net/maa/reportStatus

The controlled MAA must enter this endpoint in the Task Reporting Endpoint text box in MAA's configuration.

This endpoint must accept a POST request with Content-Type=application/json and must accept the following JSON as request content:

{
    "user":"ea6c39eb-a45f-4d82-9ecc-33a7bf2ae4dc",          // User identifier entered in MAA settings
    "device":"f7cd9682-3de9-4eef-9137-ec124ea9e9ec",        // Device identifier automatically generated by MAA
    "task":"15be4725-5bd3-443d-8ae3-0a5ae789254c",          // Task ID being reported, corresponding to ID from task retrieval
    "status":"SUCCESS",                                     // Task execution result, SUCCESS or FAILED. Generally returns SUCCESS regardless of task execution success, FAILED only in special cases noted in task descriptions
    "payload":"",                                           // Data included in report, string type. Depends on task type, e.g., screenshot task includes Base64 string of screenshot
    ...     // If your endpoint has other uses, you can add optional parameters, but MAA only transmits user and device
}

This endpoint's response content is arbitrary, but if not returning 200 OK, a notification will appear in MAA displaying "Upload failed"

Example Workflow - Controlling MAA with QQ Bot

Developer A wants to control MAA with their QQ Bot, so they developed a backend exposed on public internet providing two endpoints:

https://myqqbot.com/maa/getTask
https://myqqbot.com/maa/reportStatus

For user convenience, their getTask interface returns 200 OK and an empty tasks list regardless of received parameters. Each time they receive a request, they check the database for duplicate devices, and if none found, record the device and user in the database. In this workflow, this interface also serves as user registration.

They provide a command on their QQ Bot for users to submit deviceId.

In the QQ Bot's usage instructions, they tell users to enter their QQ number in MAA's User Identifier field and send their Device Identifier to the Bot via QQ chat.

Upon receiving the identifier, the QQ Bot checks the database for corresponding data based on the message's QQ number, instructing the user to configure MAA if not found.

Since MAA continuously sends requests once configured, if the user has properly configured MAA, matching records should exist in the database when they submit via QQ.

The Bot then marks the database record as verified, so future getTask requests with this device and user combination will return actual task lists.

When users submit commands via QQ Bot, the Bot writes a task to the database, which getTask will return shortly. Additionally, the Bot thoughtfully adds a screenshot task after each user command.

After executing tasks, MAA calls reportStatus to report results, and the Bot sends QQ messages notifying users and displaying screenshots.

Example Workflow - Controlling MAA with Website

Developer B created a website for batch MAA management with their own user management system. Their backend is publicly accessible with two anonymously accessible endpoints:

https://mywebsite.com/maa/getTask
https://mywebsite.com/maa/reportStatus

On the website, a MAA instance connection interface displays what Developer B calls a User Key (random string) and has a text box for entering device ID.

The website instructs users to enter their user key in MAA's User Identifier field and their Device Identifier on the website.

Only after successfully creating a MAA connection on the website will getTask return 200 OK; otherwise it returns 401 Unauthorized.

If users enter incorrect information in MAA and press the test connection button, they'll receive a connection test failure notification.

Users can issue tasks on the website, queue tasks, view screenshots, etc., with implementation similar to the QQ Bot example, using getTask and