Related Resources¶
📌 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 ⭐⭐⭐⭐⭐ |
| Understand how models learn and how they are used | Model training guide ⭐⭐⭐⭐⭐ |
| Look up an unfamiliar term | Glossary ⭐⭐⭐⭐ |
| Choose a CLI agent | CLI Agents Guide ⭐⭐⭐⭐⭐ |
| Connect an Agent to external tools | MCP / Skills Catalog ⭐⭐⭐⭐⭐ |
| Build a small project step by step | Cookbook ⭐⭐⭐⭐⭐ |
🧩 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. To separate Apps / Connectors, CLI Agents, and MCP Servers, see the CLI Agents Guide.
📚 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 ⭐⭐⭐⭐⭐ |
Find published MCP Servers; still check the maintainer, permissions, and source |
| servers ⭐⭐⭐⭐⭐ |
See educational reference implementations; they are not a production recommendation |
| skills ⭐⭐⭐⭐⭐ |
See how Agent Skill folders, instructions, and resources fit together |
| github-mcp-server ⭐⭐⭐⭐⭐ |
Learn OAuth, tool groups, and repository permissions from an official implementation |
| 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.
| Group | Project / document | Rating | Use | Limits |
|---|---|---|---|---|
| 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: the official discovery entry point for published Servers.
- punkpeye/awesome-mcp-servers: browse community Servers by category; review before installing.
- hesreallyhim/awesome-claude-code: Claude Code community resources.
- travisvn/awesome-claude-skills: Agent Skills community list.
- Canva MCP: official remote MCP; features, plans, and permissions depend on the account, so no fixed tool count is stated.
- Course list: choose courses by learning goal; a certificate is not a degree.
- Agent paradigms: compare common Agent shapes with diagrams.
✅ Next stop¶
- Unclear term: open
resources/glossary.en.md. - Want to build now: open
resources/cookbook.en.md. - Want to understand training choices: open
resources/model-training-guide.en.md. - Want to contribute or translate: read
resources/style-guide.en.mdandCONTRIBUTING.en.md.