Marking MCP
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Marking MCP
Marking MCP 通过 Chrome 扩展连接阅卷页和本机的评分服务。支持九科星与合成测试平台。MCP 和 CLI 共用同一个评分引擎,原图、评分细则、转录、分数和回执归档到本地 SQLite。
安装与合成卷验证
需要 Node.js 24、pnpm 11 和 Chrome 136 或更新版本。
pnpm install
pnpm build
pnpm cli init-fixture ./local-fixture/config.json 32158
pnpm cli daemon ./local-fixture/config.json在 Chrome 中打开 chrome://extensions,启用「开发者模式」,选择「加载已解压的扩展程序」,导入 packages/extension/dist/。升级时重新加载扩展并重启本地服务,以匹配 WebSocket 协议版本。从扩展弹窗打开侧栏并连接端口 32158,再打开 http://127.0.0.1:32158/fixture/marking。侧栏会显示租约、写入状态和日志。点击「检查当前标签页」可读取任务绑定、评分栏和图像来源。绑定和评分栏通过校验后可复制 JSON,切页或检查失败会清除旧结果。暂停会释放租约,恢复后自动寻找符合配置的标签页。
pnpm cli current ./local-fixture/config.jsoncurrent 返回证据图片的本地路径。每次读取写入独立的 archive/current/paper-*/ 目录。
Related MCP server: Study Errorbook
真实阅卷与自动连接
复制 真实配置模板 和 评分细则模板 到本地任务目录,按侧栏检查结果填写 origin、imageOrigins、binding 和各条 criteria[].unitId。九科星使用 https://marking.jkxjxw.com/marking 或 https://jxp.jkxjxw.com/exam-marking/marking。如果配置包含新的 HTTPS 图片来源,使用侧栏按钮授予相应站点权限。扩展核对配置的图像来源和当前文档实际加载的图片。同源图像在隔离世界保留阅卷会话凭据,跨源下载由扩展 worker 执行并省略凭据。图片字节保留在扩展侧,服务端复核格式、大小和哈希。原图请求禁止跟随重定向,每卷最多 16 MiB 和 16 张。
默认只读。获得真实提交授权后才设置 allowLiveWrites: true。使用旧配置做只读检查时,可给 serve 或 daemon 加 --read-only,覆盖配置里的写入许可。页面应使用键盘打分,扩展会在身份与分数通过提交预检后尝试关闭原生自动提交,再核验关闭状态。服务只监听 127.0.0.1,WebSocket 校验扩展的固定 Origin,并只给一份匹配的标签页租约。
MCP 客户端使用下列 stdio 配置,将路径替换为本机的绝对路径。一个配置端口使用 serve 或 daemon 中的一个,两者都提供本地 CLI 控制接口。
{
"command": "node",
"args": ["/path/to/marking-mcp/packages/server/dist/cli.mjs", "serve", "/path/to/config.json"]
}本地 CLI 会向服务核对任务键、归档绝对路径、细则哈希及调用方写许可。配置错配时返回 CONTROL_IDENTITY_MISMATCH。只读 CLI 向可写服务请求提交或刷新时返回 CONTROL_WRITE_MISMATCH。
评分前读取 阅卷协议。MCP 初始化直接加载该文件。CLI 操作如下。
命令 | 用途 |
| 读取并归档当前卷,输出原图路径。设置 |
| 调用工具。有成功回执但下一卷未就绪时,只重取下一卷。 |
| 停止服务并在交互终端确认平台保存记录。 |
| 导出本题历史,默认包含初评与复评,输出到归档目录的 |
回执与归档
submit_score_and_next 需要 paperToken、rubricSha256、transcription 和每项评分及理由。每个原生评分栏的细则满分之和必须与平台吻合。复评从最旧答卷开始,提交时必填 reviewed: true 和 reviewLimit,可用 reviewSkippedPositions 指定得到明确授权的跳过位置。位置从最旧答卷按 1 计数,reviewLimit 仍是最后位置。遇到跳过位置时,在已阅列表手动定位到下一份要评分的答卷,并保持本批提交的跳过参数一致。扩展会核对已阅列表、原分、记录 ID 和提交数据。
服务先将 pending 持久化,再调用平台。成功后以已归档的 receipt.status === "submitted" 为准。同卷重试核对扫描卷、细则、图像哈希、原评分栏及原提交证据,复评还核对原分、位置和总数。
人工对账时,独立核对平台实际保存记录和各评分栏分数。确认文件包含 archiveId、paperId、operator、platformReference 和 confirmedUnitScores,分数格式是 { "unitId": "13", "score": 4 } 对象的数组。有下一卷时,停在该卷并填写 nextPaperId,对账后需重新读取这张卷才可继续。确认末卷已保存且任务结束时填写 taskCompleted: true。两个字段只能填一个。终端会要求输入已保存的答卷 ID。使用复评归档时,命令末尾加 reviewed。
旧配置和旧细则可以直接使用。缺省 platform 时,live 默认为 jiukexing,fixture 默认为 fixture。旧 token 字段仍能读取。旧任务键、captures、images、decisions、decision_events 与恢复记录结构沿用,细则的 SHA-256 始终按原文件字节计算。HTML 按 unitId 展示分数,包含事件、图像和回执,可以离线打开。
开发
PLAYWRIGHT_SKIP_BROWSER_GC=1 pnpm exec playwright install chromium
pnpm check
pnpm build
pnpm testpackages/protocol 定义共享契约和消息。新增平台时在 packages/adapters 实现服务端驱动和 MAIN-world 适配器,将两者加入 src/registry.ts。新平台使用唯一的 id 作为归档命名空间。构建会从注册项生成 Chrome 主机权限,worker 仅对精确匹配阅卷路由的顶层文档执行适配器。packages/server 负责评分、归档和本地服务。packages/extension 负责租约、受限下载和侧栏。
本项目使用 GPL-3.0-only。九科星适配器的上游作者和版本见 NOTICE。
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Pay-per-call AI evaluation MCP server. Score LLM outputs against benchmark rubrics via Workers AI.
Remote MCP server for deterministic educational practice-assessment score conversions.
Documents to Markdown/PDF, OCR, video and audio transcripts, TTS, screenshots, domains, frames.
A search engine that learns from your grades. Store text, records, images or embeddings.
Related MCP Servers
- AlicenseBqualityCmaintenanceEnables natural-language, read-only access to a user's own learning-management account for courses, assignments and deadlines, lecture attendance, notices, grades, and syllabus evidence, including reading page/slide/cell locations inside PDF, Office, HWP/HWPX, HTML, and ZIP attachments. Runs locally over stdio or remotely over OAuth-protected Streamable HTTP for desktop and mobile clients, and can compare changes against a locally encrypted baseline.19MIT
- AlicenseAqualityCmaintenanceEnables AI hosts to import PDFs, photos, and exam papers through local stdio MCP tools, using TextIn OCR to build a versioned wrong-answer bank that keeps original evidence and image crops for human review. Supports retrieval, revision history, practice attempts, dated review queues, and printable offline HTML exports of error notebooks, student workbooks, and answer books.32MIT
- AlicenseAqualityCmaintenanceEnables users to search XJTLU past exam papers by course code, read details such as academic year, semester, and regular versus resit type, and save the chosen PDFs to local folders after logging in with their own school account. Runs as a local stdio MCP server for Codex and DeepSeek Harness, returning metadata and file paths only.516 npmMIT
- AlicenseAqualityCmaintenanceEnables users to patrol QingSoft U+ courses, track homework and experiments, read objective questions into structured briefs for a connected MCP client model to answer, fill and verify selections, and optionally submit, while handing subjective, programming, report and file work back to the human.9MIT