huly-mcp
An MCP server that enables AI assistants to interact with Huly's project and document management platform, providing search and write actions for issues, milestones, and more.
README
Huly MCP Server
This is a Model Context Protocol (MCP) server for the Huly project and document management platform, allowing AI assistants such as Claude Code and Codex to interact with Huly directly.
What is MCP?
Model Context Protocol (MCP) allows AI assistants to access external tools through a standardized interface. This server implements the MCP specification and talks to Huly through its official @hcengineering client SDK over WebSocket. It runs locally as a stdio server — the AI assistant launches it and communicates over standard input/output. There is no HTTP endpoint to host or expose.
Features
This MCP server provides access to the following Huly functionalities.
Search Actions
- List Projects: Retrieve all tracker projects and their identifiers.
- Search Issues: Find issues in a project with rich filters — title, assignee, unassigned, status, priority, component, milestone, label, parent (sub-issues), modified-after, and due-before.
- Get Issue: Fetch full detail of a single issue: markdown description, labels, parent/sub-issues, and comment & attachment counts.
- List Persons: Fetch workspace people, filtered by name.
- List Statuses: List workflow statuses (Todo, In Progress, Done…).
- List Components / Milestones / Labels: List a project's components, milestones, and available labels.
- List Comments / Attachments: Read an issue's comments (markdown, with author names) and its attachments.
Write Actions
- Create Issue: Create a new issue under a project with title, markdown description, assignee, priority, component, milestone, due date, status, and optional parent (sub-issue).
- Update Issue: Change any field of an issue — title, description, assignee, status, priority, component, milestone, due date, or estimation.
- Delete Issue: Permanently remove an issue.
- Add Comment: Post a markdown comment on an issue.
- Add Attachment: Upload a local file as an attachment on an issue.
- Add / Remove Label: Attach or remove a label (auto-creates the tag if it doesn't exist).
- Create Milestone: Create a new milestone inside a project.
Installation
# Clone the repository
git clone <REPO_URL> huly-mcp-server
# Navigate to the project directory
cd huly-mcp-server
# Install dependencies
npm install
No build step is required — the server is plain ES-module JavaScript and runs directly on Node.js (≥ 18).
Usage
Configuration
The server is configured with environment variables (or a local .env file — copy .env.example to .env):
| Variable | Description | Default |
|---|---|---|
HULY_URL |
Base URL of the Huly instance | https://your-huly-instance.com |
HULY_WORKSPACE |
Workspace identifier | your-workspace-name |
HULY_TOKEN |
Personal workspace token (recommended) | — |
HULY_EMAIL / HULY_PASSWORD |
Login used only as a fallback to mint a token | — |
Getting your token
You authenticate with a personal token so your password never lives in any config file:
cp .env.example .env # then fill in HULY_EMAIL and HULY_PASSWORD
npm run token # prints your personal HULY_TOKEN
Paste the printed token into .env as HULY_TOKEN=... and into your AI assistant's MCP config (below).
Running the server
The server is normally launched by your AI assistant, but you can start it manually to verify it connects:
npm start
Setup in Claude Code
claude mcp add huly \
--env HULY_URL=https://your-huly-instance.com \
--env HULY_WORKSPACE=your-workspace-name \
--env HULY_TOKEN=PASTE_YOUR_TOKEN_HERE \
-- node /absolute/path/to/huly-mcp-server/server.js
Or add it manually to ~/.claude.json under mcpServers:
{
"mcpServers": {
"huly": {
"type": "stdio",
"command": "node",
"args": ["/absolute/path/to/huly-mcp-server/server.js"],
"env": {
"HULY_URL": "https://your-huly-instance.com",
"HULY_WORKSPACE": "your-workspace-name",
"HULY_TOKEN": "PASTE_YOUR_TOKEN_HERE"
}
}
}
}
Restart Claude Code, then verify with claude mcp list.
Setup in Codex
codex mcp add huly \
--env HULY_URL=https://your-huly-instance.com \
--env HULY_WORKSPACE=your-workspace-name \
--env HULY_TOKEN=PASTE_YOUR_TOKEN_HERE \
-- node /absolute/path/to/huly-mcp-server/server.js
Or add it manually to ~/.codex/config.toml:
[mcp_servers.huly]
command = "node"
args = ["/absolute/path/to/huly-mcp-server/server.js"]
[mcp_servers.huly.env]
HULY_URL = "https://your-huly-instance.com"
HULY_WORKSPACE = "your-workspace-name"
HULY_TOKEN = "PASTE_YOUR_TOKEN_HERE"
Restart Codex, then verify with codex mcp list.
Development
Project Structure
huly-mcp-server/
├── server.js # MCP server — all tool definitions and Huly client wiring
├── get-token.js # Mints a personal workspace token from email + password
├── .env.example # Configuration template (copy to .env)
├── .gitignore # Excludes .env, tokens, and node_modules
├── package.json # Pinned dependencies and npm scripts
└── README.md
Each tool is defined in server.js as a server.tool(name, description, schema, handler) block. Handlers are wrapped in a guard() helper so thrown errors become clean MCP error responses instead of crashing the process.
Notes for maintainers
Discovered against our Huly instance (server 0.7.242, client 0.7.423):
- Do not use the SDK
markdown()helper withcreateDoc/updateDoc/addCollection. In client0.7.423,processMarkupassigns the upload promise without awaiting it, storing{}instead of a markup ref. Callawait client.uploadMarkup(...)explicitly and pass the returned ref. - Comment authors:
createdByis a social id — resolve viacontact.class.SocialIdentity→Person, notpersonUuid. - Person names are stored
"LastName,FirstName"— match by substring.
Testing
Run the server manually and confirm it connects and lists tools:
npm start
There is no automated test suite yet. Contributions adding one are welcome.
Contributing
Contributions are welcome! Please feel free to submit a Pull Request.
License
This project is licensed under the MIT License.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。