# Related Resources > [繁體中文](./RESOURCES.md) | [简体中文](./RESOURCES.zh-Hans.md) | **English** > [← Back to main README](README.en.md) ## 📌 Choose what you want to do now You do not need to read the whole list. Pick the task you want to do now:
| I want to… | Start here | |:---|:---| | Learn AI agents from the beginning | [Stage 0](stages/00-foundations.en.md)
⭐⭐⭐⭐⭐ | | Understand how models learn and how they are used | [Model training guide](resources/model-training-guide.en.md)
⭐⭐⭐⭐⭐ | | Look up an unfamiliar term | [Glossary](resources/glossary.en.md)
⭐⭐⭐⭐ | | Choose a CLI agent | [CLI Agents Guide](resources/cli-agents-guide.en.md)
⭐⭐⭐⭐⭐ | | Connect an Agent to external tools | [MCP / Skills Catalog](resources/mcp-skills-catalog.en.md)
⭐⭐⭐⭐⭐ | | Build a small project step by step | [Cookbook](resources/cookbook.en.md)
⭐⭐⭐⭐⭐ |
## 🧩 Three core terms: MCP, Skill, and Plugin **MCP (Model Context Protocol)**: an open protocol that lets AI applications connect to external data and tools through a shared method. What it can do still depends on the server, account permissions, and user approval. **Skill**: a reusable package of instructions that may also include scripts, templates, and references. Different products place and load Skills differently. **Plugin**: an installable package provided by a host. It may bundle Skills, commands, hooks, or MCP configuration; Plugins are not part of the MCP specification. For a hands-on example, go to [Stage 5](stages/05-claude-code-ecosystem.en.md). To separate Apps / Connectors, CLI Agents, and MCP Servers, see the [CLI Agents Guide](resources/cli-agents-guide.en.md). ## 📚 Five safe starting points Start with official entry points or complete learning materials, then evaluate community projects. These are editorial ratings, not GitHub stars.
| Starting point | Learn first | |:---|:---| | [Official MCP Registry](https://registry.modelcontextprotocol.io/)
⭐⭐⭐⭐⭐ | Find published MCP Servers; still check the maintainer, permissions, and source | | [servers](https://github.com/modelcontextprotocol/servers)
⭐⭐⭐⭐⭐ | See educational reference implementations; they are not a production recommendation | | [skills](https://github.com/anthropics/skills)
⭐⭐⭐⭐⭐ | See how Agent Skill folders, instructions, and resources fit together | | [github-mcp-server](https://github.com/github/github-mcp-server)
⭐⭐⭐⭐⭐ | Learn OAuth, tool groups, and repository permissions from an official implementation | | [hello-agents](https://github.com/datawhalechina/hello-agents)
⭐⭐⭐⭐⭐ | Learn Agent principles and implementation through a Chinese-language course |
> ⚠️ **Start with fewer keys**: an MCP Server may read, create, send, or delete real data. Begin with read-only access, least privilege, and a test workspace; keep **Human Approval** before write, send, or delete actions. ## 🔌 Common integration groups The table below shows selected projects and official entry points directly. Start with the group you need; only the longer supplemental list is collapsed.
GroupProject / documentRatingUseLimits
Notes / Knowledge Bases⭐⭐⭐⭐⭐
Use: Search, create, and update Notion content
Status / limits: Official hosted MCP; user OAuth required and permissions follow the signed-in user
⭐⭐⭐⭐
Use: Read and write an Obsidian vault
Status / limits: Community project; grant only the vault and write scope needed
⭐⭐⭐⭐⭐
Use: Summarize and ask questions about your own sources
Status / limits: Google service; availability varies by region, account, and feature
⭐⭐⭐⭐
Use: Operate Gemini Notebook through Python / CLI
Status / limits: Unofficial interface; Google changes may break it
Office Documents⭐⭐⭐⭐⭐
Use: Read Agent Skill examples and document workflows
Status / limits: Official examples; built-in Skills vary by Claude surface
⭐⭐⭐⭐
Use: Extend Office document automation workflows
Status / limits: Community Skills; back up files and verify outputs
⭐⭐⭐⭐
Use: Connect Gmail, Drive, Docs, Sheets, Slides, Calendar, and Chat
Status / limits: Official Developer Preview; each product has a dedicated server, with OAuth 2.0 and user / organization governance
Dev teamwork⭐⭐⭐⭐⭐
Use: Query and work with issues, PRs, and repositories
Status / limits: Official MCP; use OAuth or a least-privilege token and approve high-impact writes
⭐⭐⭐⭐⭐
Use: Search or update Jira, Confluence, and Bitbucket
Status / limits: Official hosted remote MCP; OAuth 2.1 and signed-in user permissions apply; require human approval before writing real data
⭐⭐⭐⭐⭐
Use: Search and update Linear issues / projects
Status / limits: Official hosted remote MCP over Streamable HTTP; an official read-only option exists
⭐⭐⭐⭐⭐
Use: Search Slack, send messages, and manage canvases
Status / limits: Official MCP; it is not read-only, so confirm before sending or changing content
Research Workflow⭐⭐⭐⭐⭐
Use: Turn research steps into reusable Skills
Status / limits: Community project; install only Skills relevant to the current task
⭐⭐⭐⭐⭐
Use: Connect Zotero, Obsidian, and research workflows
Status / limits: Community workspace; check data location, backups, and permissions
⭐⭐⭐⭐
Use: Organize Zotero data with Skills
Status / limits: Community Skill; back up the library before batch changes
⭐⭐⭐⭐
Use: Delegate clearly scoped repetitive work to another Agent
Status / limits: Community tool; lock files, acceptance criteria, and stop conditions first
Chinese Ecosystem⭐⭐⭐⭐
Use: Convert between Feishu (Lark) documents and Markdown
Status / limits: Community tool; confirm sharing and write permissions
Expand more lists, courses, and design tools - [MCP Registry](https://registry.modelcontextprotocol.io/): the official discovery entry point for published Servers. - [punkpeye/awesome-mcp-servers](https://github.com/punkpeye/awesome-mcp-servers): browse community Servers by category; review before installing. - [hesreallyhim/awesome-claude-code](https://github.com/hesreallyhim/awesome-claude-code): Claude Code community resources. - [travisvn/awesome-claude-skills](https://github.com/travisvn/awesome-claude-skills): Agent Skills community list. - [Canva MCP](https://www.canva.dev/docs/mcp/): official remote MCP; features, plans, and permissions depend on the account, so no fixed tool count is stated. - [Course list](resources/courses.en.md): choose courses by learning goal; a certificate is not a degree. - [Agent paradigms](resources/agent-paradigms.en.md): compare common Agent shapes with diagrams.
## ✅ Next stop - Unclear term: open [`resources/glossary.en.md`](resources/glossary.en.md). - Want to build now: open [`resources/cookbook.en.md`](resources/cookbook.en.md). - Want to understand training choices: open [`resources/model-training-guide.en.md`](resources/model-training-guide.en.md). - Want to contribute or translate: read [`resources/style-guide.en.md`](resources/style-guide.en.md) and [`CONTRIBUTING.en.md`](CONTRIBUTING.en.md).