The protocol connecting AI to tools
MCP is the open protocol connecting AI models to external tools and data — a standard interface between clients and servers, so any AI client can use any tool.
Updated 2026-08-15. Same-day snapshot of the official repo.
How to Get Started with MCP
3 steps to understand the protocol and connect your first server.
Read the spec
Understand the client-server model and the three primitives.
https://modelcontextprotocol.ioInstall a server
Pick one from the official servers repo (e.g. filesystem) and register it in your client.
claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem /pathConnect your data
Write your own server (TS SDK or Python) to expose business tools to the model.
npx @modelcontextprotocol/create-serverPitfalls
License transition
MCP is mid-transition from MIT to Apache-2.0; check per-repo/file licensing before commercial use.
Fragmented ecosystem
Servers, SDKs and registries spread across repos; third-party server quality varies. Prefer official servers and reviewed registry entries in production.
Security surface
Connecting tools gives the model more power and more risk. Use read-only DB accounts; require approval for writes.
Beside Agent Skills & plugins
MCP is the protocol layer, Skills the knowledge layer, plugins the product layer. This site covers all three so you pick the right abstraction.
| Tool | Nature | Problem solved | License | Typical user |
|---|---|---|---|---|
| MCP | Open protocol | Standard interface for AI-to-tool/data | MIT → Apache-2.0 | Developers / tool vendors |
| Agent Skills | Procedural knowledge format | Teach the model how to work | Varies by repo | Agent users / teams |
| 插件系统（各家） | Product-internal extension | Add capabilities to one tool | Varies by product | Power users of one tool |
Official Directory
Official
- Official site
Specification, docs and getting started.
- Servers repository
Reference servers: filesystem, search, databases (89.6k★).
Spec & SDK
- Specification
Protocol specification and docs (8.9k★).
- TypeScript SDK
Official TS SDK for servers and clients (13.2k★).
- MCP Registry
Official discoverable server registry.
Community
- Discord
Official MCP Discord.
Host clients
- Claude Code (client)
A flagship MCP host client, covered in its own hub.
Latest Updates
2026-08-14 · GitHub · modelcontextprotocol
MCP transitions from MIT to Apache-2.0MCP moves from Anthropic-led to open governance, relicensing from MIT to Apache-2.0 for new contributions. The servers repo passes 89.6k stars as the de-facto standard connecting AI to tools.
LaoJin's Pick
FAQ
- What is MCP?
- MCP is the open protocol connecting AI models to external tools and data. It defines a standard interface between clients (e.g. Claude Code) and servers (exposing tools/resources), so any AI client can use any tool without pairwise integration.
- What are MCP's core concepts?
- Three primitives: Tools (functions the model calls), Resources (data/files to read), Prompts (reusable templates). Servers expose them; clients consume them.
- How does MCP differ from Agent Skills?
- MCP is a runtime protocol for connecting tools/data; Agent Skills are procedural knowledge packages teaching the model how to work. They complement each other; Claude Code supports both.
- What is MCP's license?
- Transitioning from MIT to Apache-2.0 for new code and spec contributions; docs are CC-BY-4.0. Older MIT contributions remain MIT until relicensed.
Need to deploy a team-level Agent system?
Open-source frameworks are free, but turning a runtime into an enterprise asset requires access control, RBAC sandboxes, and private RAG integrations. Laojin Global provides commercial-grade architecture and deployment support.
Cite This Hub
We encourage you to capture this hub into your knowledge base, research, or feed it straight to an AI agent under CC BY 4.0.
How this hub stays current
Version and links live in src/content/mcp.ts; star snapshots sync via GitHub API.