Most people talking about the MCP ecosystem fixate on OpenAI, Anthropic, and the major overseas players. What many haven't noticed: Xiaohongshu's Skill ecosystem quietly built itself up. Reportedly 160,000 developers and 7,300 Skills. Domestically, that's already top-tier.
This post breaks down Xiaohongshu's Skill ecosystem: what's the business logic, what does it mean for MCP, and what can agent platform builders and developers learn from it.
First: What Is Xiaohongshu's Skill?
Short version: Xiaohongshu's Skill is essentially the same thing as MCP—a standardized tool-calling protocol. Developers follow the protocol, package their capability as a Skill, and publish it to Xiaohongshu's platform. Users calling Xiaohongshu's AI assistant can invoke these Skills.
For example:
- A developer with hotel data builds a hotel lookup Skill and publishes it
- A developer with travel guide content builds a guide generation Skill
- A developer with flight data builds a flight lookup Skill
User tells Xiaohongshu's AI assistant "find me a hotel in Shanghai." The AI calls the hotel lookup Skill and returns results. The user doesn't know which developer built the Skill behind the scenes.
See? This model is identical to MCP. Build a standardized protocol. Developers package capabilities as tools. Platform distributes them. Xiaohongshu just calls it "Skill."
160K Developers and 7,300 Skills: Where Does That Rank?
Scale first. 160K developers, 7,300 Skills. How big is that?
Horizontal comparison:
- MCP servers on GitHub number in the thousands
- Domestic MCP ecosystems mostly sit at a few hundred servers
- Xiaohongshu at 7,300 Skills is already top-tier domestically
More importantly: developer count. 160K is a lot. Most platforms building ecosystems have a few thousand or tens of thousands of developers. Xiaohongshu hit 160K, which says something about ecosystem pull.
Why did Xiaohongshu pull this off? I see a few reasons:
First, it has traffic. Xiaohongshu already has hundreds of millions of users. Developers don't worry about who uses their Skill. Publish one and users start calling it immediately.
Second, it has scenarios. Xiaohongshu is a lifestyle platform. Travel, food, fashion, beauty—these scenarios naturally fit Skills. Users already search for this content on the platform.
Third, it has monetization. Developers make money on the platform. Money attracts developers.
What's the Business Logic?
What's Xiaohongshu's business logic behind building this Skill ecosystem? I see three layers:
Layer 1: Boost the AI assistant's competitiveness
Everyone is building AI assistants. Xiaohongshu needs one too. How do you make yours better? Through ecosystem. Other assistants can chat. Yours can call various Skills and get things done. That's the edge.
Layer 2: Keep users in the Xiaohongshu ecosystem
Before, users read a guide on Xiaohongshu and wanted to book a hotel, so they jumped to Ctrip. Now with Skills, users book directly in Xiaohongshu. Traffic doesn't leak out. Transactions happen on the platform.
Layer 3: Earn transaction commission
Users book hotels and buy products through Skills. Xiaohongshu takes a commission. That's the endgame—evolve from a content platform into a content + transaction platform. AI assistants and Skills are the tools for that upgrade.
Plain version: Xiaohongshu isn't building the Skill ecosystem out of charity. It's upgrading from "content community" to "content + transaction platform."
Developer Incentives: How Does It Attract Developers?
The hardest part of building an ecosystem isn't the protocol or the platform. It's getting developers to show up. Why would they come? Because they make money.
Xiaohongshu's developer incentives have a few things I think are well done:
1. Pay per call
Your Skill gets called N times, you get paid per call. Developers don't worry about monetization. If your Skill is useful and call volume grows, money follows. Friendly to small developers.
2. Traffic support
Good Skills get platform traffic support—featured on the homepage, more exposure. Developers build a good Skill and get more calls. Positive loop.
3. Low barrier
Building a Skill has a low barrier. No need for complex tech. Follow the docs. Individual developers and small teams can do it. That's why it hit 160K developers—the barrier is low enough for everyone to join.
Compare with MCP ecosystems. Many MCP platforms let developers build servers, but there's no monetization path. Developers build, earn nothing, and stop. Xiaohongshu's pay-per-call model solves this. Developers build and get paid directly.
What Does This Mean for the MCP Ecosystem?
Xiaohongshu's Skill ecosystem has a few important lessons for the broader MCP ecosystem:
First, ecosystem growth requires solving developer monetization
Many MCP platforms only build protocol, directory, and showcase. They don't solve how developers make money. Result: developers build a bunch of MCP servers, earn nothing, and gradually stop. Xiaohongshu's pay-per-call model fixes this.
Second, ecosystem needs real scenarios, not empty structure
Many MCP ecosystems build tools without knowing where they're used. Xiaohongshu's strength: it's already a content platform with real users and real scenarios. Skills aren't built for the sake of building. They solve real user needs.
Third, platform needs traffic for developers to show up
Developers build Skills because people will use them. If the platform has no traffic, developers build Skills and no one calls. Why would they show up? Xiaohongshu has hundreds of millions of users. Publish a Skill and traffic arrives immediately. That's the biggest draw.
RollingGo's hotel MCP has close to 300 GitHub stars, with both an MCP server and travel Skills. Supports 40+ major LLM clients directly. Behind it: 2M+ global hotel properties and 110K+ direct-contracted hotels. Direct inventory with real-time price confirmation. What you see is what you can book. Completely free, no call-volume limits. Provides an all-in-one booking skill for ClawHub and Coze. Developer partners set country-based markup rates and earn commission.
This is the same logic. Developers build AI agents and need hotel data, so they connect your MCP. Platform and developers need each other. Want to see how this developer ecosystem works? Follow the quick-start guide and get a key.
Setup in Manus is three fields:
Server name: RollingGo-Hotel
Server URL: https://mcp.rollinggo.ai/mcp
Auth: Bearer YOUR_API_KEY
Save, and you're calling hotel search, room detail, and price confirmation in Manus chat. MCP and Skills are like bottom-layer data and upper-layer applications. They need each other.
Future: Which Agent Ecosystem Has the Most Promise?
Lots of platforms building agent ecosystems now. OpenAI GPTs, Anthropic MCP, ByteDance Coze, Xiaohongshu Skills, various open-source platforms. Which one has the most promise?
My read: platforms with traffic, scenarios, and monetization all three have the most promise.
Only tech, no traffic or scenarios, can't grow. Only traffic, no monetization, developers leave. Only monetization, no scenarios, tools are empty.
From this angle, Xiaohongshu has an edge—traffic (hundreds of millions of users), scenarios (lifestyle content), monetization (transaction commission). Its Skill ecosystem could become one of China's largest agent ecosystems.
Still early. It's not decided yet. But Xiaohongshu's case is worth studying for every company building an agent ecosystem.
Xiaohongshu's Skill ecosystem quietly reached 160K developers and 7,300 Skills. Many haven't noticed. Behind the number: Xiaohongshu's ambition to upgrade from content platform to transaction platform.
For developers, this is also an opportunity. A new ecosystem is rising, and new dividends come with it. Building a Skill now might have less competition and more opportunity than building an MCP server.
Which platform's agent ecosystem do you think has the most promise? Xiaohongshu, ByteDance, or OpenAI abroad? Drop a comment.
Top comments (0)