Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@mcp-personal search for papers on graph neural networks"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A personal MCP server built with Python (FastMCP) that can connect to any MCP client such as Cursor / VS Code / DeepSeek Harness.
See PLAN.md for the full plan · See docs/展示指南.md for demo materials · Current progress: four scenarios + HTTP + auth + pdf_extract ✅ · 21 tests passing · See .env.example for environment variables
# 1. 创建虚拟环境并安装（需 Python 3.11+ 或 uv）
uv venv
uv pip install -e ".[dev]"
# 2. 跑测试
uv run pytest tests -q
# 3. 启动服务器（stdio，供客户端接入）
uv run python -m mcp_personal
# 4. 冒烟测试：真实 stdio 子进程链路
uv run python scripts/smoke_stdio.py
# 5. 查看服务器注册的能力清单（Tools/Prompts/Resources）
uv run python scripts/verify_capabilities.py
# 6. 调试：官方 MCP Inspector
uv run mcp dev src/mcp_personal/server.py
Related MCP server: DDG MCP Server
Configuration is preconfigured (portable approach: uv run --project . automatically locates the environment, so the repository can be moved anywhere):
Client | Config file |
Cursor |
|
VS Code |
|
Other | Standard stdio config: |
After connecting, have the AI call the ping tool to verify the link.
# 终端 1：启动 HTTP 服务（默认 127.0.0.1:8000，--host 0.0.0.0 可局域网访问）
uv run python scripts/run_http.py --port 8000
# 终端 2：用官方客户端验证链路
uv run python scripts/client_http_check.py --url http://127.0.0.1:8000/mcpFor other remote clients, simply set the MCP server address to http://<host>:<port>/mcp.
After setting the environment variable and starting, all requests must include Authorization: Bearer <token>:
# 终端 1：带鉴权启动
MCP_API_TOKEN=my-secret uv run python scripts/run_http.py --port 8000
# 终端 2：带 token 验证
uv run python scripts/client_http_check.py --url http://127.0.0.1:8000/mcp --token my-secretNote: Based on FastMCP's official
StaticTokenVerifier(API key style), the token is stored in plaintext and is only suitable for personal/development use; production should switch to OAuth/JWT.
├── PLAN.md # 项目规划书
├── pyproject.toml # 依赖与打包
├── src/mcp_personal/ # 服务器代码
│ ├── server.py # FastMCP 入口（注册全部原语）
│ ├── tools/ # Tool：paper(检索)/dataset(摸底)/scaffold(课设)/devtools(目录树)
│ ├── prompts.py # Prompt：literature_review 文献综述框架
│ └── resources.py # Resource：papers://list 本地论文库
├── papers/ # 本地论文库（放 PDF）
├── datasets/ # 竞赛数据集（dataset_info 工具使用）
├── tests/ # pytest 测试（13 个）
├── .github/workflows/ # CI：push/PR 自动跑测试
├── scripts/ # smoke_stdio.py / verify_capabilities.py
├── .cursor/ .vscode/ # 客户端接入配置
└── README.md生成一份课设报告 Markdown 骨架并保存到 reports/ 目录，返回保存路径与结构。
title: 报告标题（同时作为文件名）
course_name: 课程名称
student: 姓名学号（默认占位）
sections: 主体章节数（1-8）
Inputs: title (required), course_name (required), student, sections. Returns: result
分析本地数据集（CSV/JSON），返回结构/行数/每列统计，供竞赛数据处理前快速摸底。
path: 数据集路径；相对路径默认基于 datasets/ 目录（或 MCP_DATASETS_DIR），绝对路径也可
Inputs: path (required). Returns: object
原样返回输入的文本，用于测试工具参数传递是否正常。
Inputs: text (required). Returns: result
按关键词在 arXiv 检索学术论文，返回结构化列表（标题/作者/年份/摘要/链接）。
query: 检索关键词，如 "model context protocol"
max_results: 最多返回条数（1-20，默认 5）
year_from: 起始年份（含），如 2023；留空不限
category: arXiv 分类，如 cs.AI、cs.SE；留空不限
sort_by: 排序方式，relevance（相关度）或 date（最新优先）
Inputs: query (required), sort_by, category, year_from, +1 more. Returns: result
下载 arXiv 论文 PDF 并解析正文，返回页数与正文预览（论文精读前快速浏览用）。
arxiv_id: arXiv 编号，如 "2507.19570v1"
max_chars: 预览字符数上限（100-5000，默认 500）
Inputs: arxiv_id (required), max_chars. Returns: object
连接自检：返回 "pong"。客户端接入后先调用它确认链路通畅。
Inputs: no named parameters. Returns: result
列出目录的文件树（只读），默认跳过 .venv/.git/pycache 等目录。
path: 目录路径（默认当前目录；可用绝对路径）
max_depth: 最大递归深度（1-4）
skip_hidden: 跳过隐藏目录与常见依赖目录
Inputs: path, max_depth, skip_hidden. Returns: result
创建一个新项目的目录骨架（README/.gitignore/入口文件），返回创建的文件清单。
name: 项目名（单级目录名，防路径穿越）
language: python / node / plain
description: 项目简介，写入 README
Inputs: name (required), language, description. Returns: result
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 8 tools
The eight actual tools have distinct purposes: connectivity/test, arXiv search, PDF text preview, dataset stats, project scaffolding, report generation, and file-tree listing. ping and echo are both test helpers but differ clearly (link check vs parameter passthrough), so an agent is unlikely to select the wrong tool.
Naming is a mix of bare verbs (ping, echo), verb+object (scaffold_project), object+verb (pdf_extract), and noun compounds (paper_search, project_tree, course_report). All names are lowercase, but there is no consistent verb_noun or action-oriented convention.
Eight tools is comfortably within the ideal range for a personal MCP server, and each tool provides a genuinely different capability. None feel like padding, and the count is not overwhelming for an agent to choose from.
For the inferred domain of research/development personal utilities, the advertised workflows are largely covered: search then extract papers, inspect datasets, scaffold a project, generate a report, and view a file tree. Minor gaps exist (e.g., no dataset sample-row preview or paper metadata fetch by ID), but they are workable and not severe dead ends.
A MCP server built for developers enabling Git based project management with project and personal…
Nifty's MCP server — exposes tasks, projects, messages, and files as tools for AI agents.
MCP server for accessing curated awesome list documentation
