
Ai Hr Management Toolkit
io.github.XJTLUmedia由 GitHub 账号 XJTLUmedia 发布v3.0.5更新于 Oct 24, 2026
AI HR toolkit: 24 MCP tools for resume parsing, skill extraction & ATS management.
概览
一个本地 MCP 服务器,让 AI 助手解析简历、提取技能并给候选人打分,还提供覆盖职位、候选人、面试和录用通知的完整招聘管理系统。
- 可以做什么
- 解析简历与文档:读取 PDF、DOCX、TXT、Markdown 和网址内容,输出结构化 JSON,支持单个与批量处理,并可查看分析流程各阶段与置信度。分析候选人:提取关键词、模式、实体、技能、经历,以及与职位描述的相似度。评分与排序:按加权标准评估候选人,给出通过、复核或拒绝的结论。管理招聘流程:创建和编辑职位、候选人、面试、录用通知与备注,并提供搜索、分析、评分卡、人才库、入职清单与合规报告。导出与发送:将结果导出为 JSON 或 CSV 并通过邮件发送。生成测试用示例数据。
- 适用场景
- 适合需要让助手批量筛选简历、按职位描述比较候选人,或在不另装招聘系统的情况下维护招聘流程的场景。适合在桌面 MCP 客户端中使用的招聘人员和中小型 HR 团队。
- 运行要求
- 以 npm 包形式通过 stdio 在本地运行,通常用 npx 启动,因此需要 Node.js,且工作目录中要有 package.json。算法类工具不需要 API 密钥或环境变量;如需 LLM 辅助评分,可在单次调用中提供 AI 服务商密钥。README 还介绍了自行部署的 HTTP 端点方式。
安装
在 SourceWeft 中
- 打开 控制台中的 Ai Hr Management Toolkit,将其添加到工作区。
- 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
AI HR Management Toolkit
[image] [image] [image]AI-powered resume parser & full Applicant Tracking System with 21 MCP tools. Parse PDFs, extract skills, detect patterns, score candidates, and manage a complete hiring pipeline — all from your AI assistant, no manual work required.
Live demo: https://ai-hr-management-toolkit.vercel.app
[mcp-ai-hr-management-toolkit server]What Is This?
You have 50 resumes to screen. Your AI assistant can reason about candidates — but it cannot open PDFs, extract structured data, or track pipeline stages. This toolkit bridges that gap.
Give your AI assistant 21 tools covering the entire hiring workflow:
- Parse PDFs, DOCX, TXT, Markdown, and URLs into structured JSON
- Extract skills, experience, keywords, and entities algorithmically
- Score and rank candidates against job descriptions
- Run a full ATS: jobs, candidates, interviews, offers, notes, and analytics
20 of 21 tools are 100% algorithmic — no LLM calls, no API keys required. The AI calls tools, interprets the results, and delivers analysis. You just ask questions.
Quick Start (MCP Clients)
No installation needed. Point your MCP client at the package:
Claude Desktop — Edit %APPDATA%\Claude\claude_desktop_config.json (Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):
Example usage:
[image] [image]Cursor — Add to .cursor/mcp.json in your project root:
VS Code Copilot — Create .vscode/mcp.json in your project root:
VS Code users: Run the
npxcommand from a directory that contains apackage.json(i.e. any project root). Thecwdkey in.vscode/mcp.jsoncan override the working directory if needed.
Windsurf / other MCP clients — Use the same npx pattern above.
Installation Options
Option 1: NPX (Zero-install, recommended)
Works from any project directory (requires a package.json in the working directory):
Option 2: Global install
Install once, use from any directory:
Option 3: Remote HTTP endpoint
Deploy the Next.js app and use the Streamable HTTP transport:
Test locally:
Option 4: Local development (Web UI + MCP)
Web UI at http://localhost:3000. MCP endpoint at http://localhost:3000/api/mcp. No .env needed — configure API keys in the UI or pass them per tool call.
All 21 MCP Tools
All tools return structured JSON with next_steps hints so the AI knows what to call next.
Resume Parsing & Ingestion
Unified Analysis
analyze_resumeconsolidates what were previously 7 separate tools (extract_keywords,detect_patterns,classify_entities,extract_skills_structured,extract_experience_structured,compute_similarity,analyze_resume_comprehensive) into a single entry point with aspect selection.
Candidate Matching & Scoring
Export & Notifications
ATS — Jobs
ATS — Candidates & Pipeline
ATS — Interviews
ATS — Offers & Notes
ATS — Enterprise HR
Testing & Seeding
assess_candidateoptionally calls an LLM when you supplyprovider+apiKey; it falls back to fully algorithmic scoring otherwise.
Example Multi-Turn Flow
Analysis Pipeline
Every resume runs through a 5-stage algorithmic pipeline:
- Ingestion — PDF via pdf-parse v2, DOCX via mammoth, HTML/URL via cheerio, plain text/markdown natively
- Sanitization — Removes non-ASCII artifacts, normalizes whitespace, strips formatting noise
- Tokenization — TF-IDF with unigrams, bigrams, and trigrams; scored by document frequency
- Classification — NER with domain-aware disambiguation (e.g. "Java" as language vs. Indonesian city; "Go" as language vs. verb)
- Serialization — Maps entities to typed
ResumeSchemawith confidence scores and data quality metrics
Supported File Formats
Max file size: 10 MB
Structured Output Schema
Web UI
The app ships with a full web interface:
Switch AI providers from the selector at the top. Supports OpenAI, Anthropic, Google, DeepSeek, GLM, Qwen, OpenRouter, and OpenCode Zen.
REST API Endpoints
All endpoints accept multipart/form-data with optional headers:
Tech Stack
Development
Project Structure
License
来源:README.md,提交 cbeba6c
工具
0版本历史
1- v3.0.5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