API is MCP | Higress launches MCP Marketplace to accelerate legacy APIs into the MCP era.

MCP is becoming the "outpost" for companies to obtain AI traffic, as application vendors and SaaS service providers compete to build their own MCP Servers. However, challenges also arise:

• MCP is a brand new communication protocol. Although it provides a programming language-agnostic data format like JSON-RPC, there are high server construction and debugging costs when facing new features such as "Resource, Tool, Prompts".

• Most of the MCP Servers that have been released to various markets have not yet supported Streamable HTTP, leading to transformation costs and long-term maintenance costs at the protocol level.

• Enterprise users building their own MCP market lack best practices and complete solutions.

Based on the above challenges, Higress has open-sourced "batch conversion of existing APIs to MCP Servers" and the ability of "MCP Servers deployed based on Higress to support protocol offloading, eliminating the maintenance work associated with releasing new versions of MCP" and launched the Higress MCP Marketplace.

MCP has released two versions on 2025-03-26 and 2024-11-05, with 2025-03-26 supporting Streamable HTTP. The early adopters of MCP face a heavy upgrade workload.

01 Why did Higress launch the MCP Marketplace?

Although MCP is popular, it is still in the early stages of development, and a single MCP market is difficult to meet the diverse needs of various parties. Different enterprises and developers have significant differences in technical capabilities, application scenarios, business needs, and selection, making a diversified MCP market inevitable.

Currently, the MCP market is roughly divided into four categories:

• "App Store" players: A distribution center for MCP across the network, with the largest number of Servers and Clients, and they will launch the latest resources immediately and open upload entrances, but debugging, hosting, and other value-added services are not yet provided. Customers need to deploy server resources by themselves for debugging and calls. For example, mcpmarket.com, mcp.so, etc.

• Client players: Maintained by clients, such as large model robots and AI programming, invoking officially supported Servers through natural language and code, fully reasoning during calls, resulting in better invocation and output effects for MCP. For example, Cline and Tongyi Lingma, etc.

• Platform-type players: Maintained by cloud vendors and comprehensive open-source communities, providing management services for the entire lifecycle of MCP, including debugging, invocation, release, hosting, etc., simplifying the construction process of Agent applications, divided into official hosting (Host) and local deployment (Local). For official hosting (Host), users can directly invoke and test, with computation resources provided by the platform; for local deployment (Local), users need to deploy server resources by themselves for debugging and invocation. For example, Bailian, Modai. Dify, and Spring AI Alibaba can also be classified as such, simplifying implementation and invocation methods but not providing computational resource support.

• Vertical players: Single open-source projects or cloud products or PaaS vendors maintain them, providing support for various operating environment types for MCP servers, leveraging their own professional capabilities to provide value-added services in network protocol adaptation, request routing, security, server operations, and expansion.

The Higress MCP Marketplace belongs to the fourth category mentioned above. Unlike other marketplaces, Higress primarily addresses the following three needs:

• Rapid conversion of existing APIs to MCP Servers: Users do not need to start from scratch; in other words, Higress Marketplace is aimed at enterprises with a large number of existing APIs that wish to quickly enjoy the benefits of MCP.

• Enterprise-level needs: Leverage the mature capabilities of gateways in authentication, authorization, rate limiting, and observability to meet enterprise-level MCP service requirements.

• Self-built enterprise MCP markets: Higress MCP Marketplace serves as a complete example, providing enterprises with a comprehensive MCP market solution, including rapid generation of MCP Servers, and performing authentication, traffic control, and other capabilities when providing external invocation services.

The first phase launched nearly 50 MCP Servers, all of which are implemented and maintained officially by Higress, and the implementation code has been published as best practices to help enterprises understand how Higress converts existing APIs into MCP Servers.

02 How is Higress MCP Marketplace different?

A major feature of Higress MCP Marketplace is that it is platform-agnostic, large models, and clients can all be used as long as they comply with the relevant protocols. Please refer to Q&A at the end for Q4/Q5/Q6. Additionally, it includes features such as batch conversion of existing APIs, refined tuning, and enterprise-level capabilities.

2.1 Batch Conversion of Existing APIs

Utilizing existing infrastructure has become a key strategy for most enterprises to enter the MCP era, but it introduces a new problem: converting existing REST APIs to MCP Servers is a "repetitive physical task," and daily maintenance, including interface updates and server stability assurance, is required.

To enable the AI assistant to invoke REST APIs, developers need to:

1. Write MCP server code

2. Implement adaptation logic for each API endpoint

3. Handle parameter conversion and response formatting

4. Deploy and maintain this code

This process is not only cumbersome, but code needs to be updated whenever APIs change. Higress API-to-MCP can thoroughly solve this problem. With simple declarative configuration, any REST API can be converted into an MCP Server without writing any code, and the entire process takes only a few minutes.

Let’s look at a practical example to see the powerful features of API-to-MCP. Here is the configuration for converting the Gaode map's geocoding API into an MCP tool:

server:
  name: rest-amap-server
  config:
    apiKey: your-api-key-here
tools:
- name: maps_geo
  description: "Converts detailed structured addresses into latitude and longitude coordinates. Supports parsing landmark names and building names into latitude and longitude coordinates."
  args:
  - name: address
    description: "The structured address information to be parsed."
    required: true
  - name: city
    description: "The specified city for the query."
    required: false
  requestTemplate:
    url: "https://restapi.amap.com/v3/geocode/geo?key={{.config.apiKey}}&address={{.args.address}}&city={{.args.city}}&source=ts_mcp"
    method: GET
    headers:
    - key: x-api-key
      value: "{{.config.apiKey}}"
  responseTemplate:
    body: |
      # Geocoding Information
      {{- range $index, $geo := .geocodes }}
      ## Location {{add $index 1}}
      - **Country**: {{ $geo.country }}
      - **Province**: {{ $geo.province }}
      - **City**: {{ $geo.city }}
      - **City Code**: {{ $geo.citycode }}
      - **District/County**: {{ $geo.district }}
      - **Street**: {{ $geo.street }}
      - **Number**: {{ $geo.number }}
      - **Administrative Code**: {{ $geo.adcode }}
      - **Coordinates**: {{ $geo.location }}
      - **Level**: {{ $geo.level }}
      {{- end }}

What does this configuration do?

1. Defined a tool named maps-geo for converting addresses to coordinates

2. Declared two parameters: the required address and the optional city

3. Set up a request template, inserting parameters and configuration values into the URL and request headers

4. Created a response template, converting the JSON response into a structured Markdown format

When the AI assistant invokes this tool, it will:

🔹 Use the provided address and city parameters to build the API request

🔹 Invoke the Gaode map API

🔹 Convert the JSON response into an easy-to-read Markdown format

🔹 Return the formatted result to the AI assistant

The entire process requires no coding; just a simple configuration file.

Advanced Usage: Complex JSON Handling

The true power of API-to-MCP lies in its ability to handle complex JSON responses. By using GJSON path syntax, you can perform advanced queries and transformations:

<!-- Using the gjson function for complex queries -->
Active Users: {{gjson "users.#(active==true)#.name"}}

<!-- Array filtering with multiple conditions -->
Active Developers Over 30: {{gjson "users.#(active==true && age>30)#.name"}}

<!-- Using modifiers -->
Usernames (in reverse order): {{gjson "users.@reverse.#.name"}}

<!-- Iterating and filtering results -->
Admins:
{{range $user := gjson "users.#(roles.#(==admin)>0)#"}}
  - {{$user.name}} ({{$user.age}})
{{end}}

These powerful query capabilities allow you to extract the most valuable information from complex API responses and present it in a way that's most understandable and usable by AI assistants.

2.2 Refined Tuning

Automatically generated configurations are complete but often lack refinement. Especially when APIs return complex data structures, if configurations are not manually refined, it can lead to the LLM misunderstanding the data, negatively impacting user experience. For example, it may cause the following issues:

🔹 Information overload: LLM’s context window is limited, and too much irrelevant information can dilute the important content.

🔹 Lack of clarity of structure: Complex nested relationships can be lost in flat descriptions.

🔹 Lack of semantics: Technical codes and jargon that have not been converted are difficult for LLM to understand correctly.

🔹 Hallucination risks: Faced with unfamiliar data structures, LLM may make incorrect inferences.

Higress supports combining Go Template and GJSON expressions for precise processing of request and response templates (for detailed capabilities, please refer to the documentation: https://higress.cn/en/ai/mcp-server). Through careful tuning, the above issues can be addressed. Compared to manual tuning, Higress tuning only requires configuration changes, and the process of configuration changes is completely lossless for traffic, with SSE connections remaining intact.

Let’s look at a case before and after tuning:

Before tuning (based on Higress's open-source openapi-to-mcp command line tool, generated automatically using Swagger documentation):

server:
  name: ecommerce-api
tools:
  - name: searchProducts
    description: "Search for products in the e-commerce platform"
    args:
      - name: query
        description: "Search query string"
        type: string
        required: true
      - name: category
        description: "Product category"
        type: string
        required: false
      - name: limit
        description: "Maximum number of results to return"
        type: integer
        default: 10
    requestTemplate:
      url: "https://api.example.com/products/search"
      method: GET
      argsToUrlParam: true
    responseTemplate:
      prependBody: |
        # Search Results
        
        Below is the API response with these fields:
        
        - **success**: Boolean indicating if the request was successful
        - **total**: Total number of matching products
        - **page**: Current page number
        - **pageSize**: Number of items per page
        - **products**: Array of product objects with the following fields:
          - **id**: Product unique identifier
          - **name**: Product name
          - **description**: Product description 
          - **price**: Product price
          - **compareAtPrice**: Original price before discount
          - **currency**: Currency code (e.g., USD, EUR)
          - **availability**: Product availability status
          - **metadata**: Technical metadata
          - **attributes**: Product attributes
          - **variants**: Product variations
          - **images**: Product images
          - **categories**: Categories the product belongs to
          - **tags**: Product tags
          - **brand**: Product brand information
          - **shipping**: Shipping information
          - **ratings**: Product ratings and reviews
        
        Original response:

After tuning:

server:
  name: ecommerce-api
tools:
  - name: searchProducts
    description: "Search for products on the e-commerce platform and return a list of products that match the search criteria, including basic information, prices, stock status, ratings, etc."
    args:
      - name: query
        description: "Search keywords, which can be product names, brands, models, or keywords in the description."
        type: string
        required: true
      - name: category
        description: "Product category, such as 'electronics', 'clothing', 'home', etc."
        type: string
        required: false
      - name: limit
        description: "Number of results to return, range 1-50."
        type: integer
        minimum: 1
        maximum: 50
        default: 10
    requestTemplate:
      url: "https://api.example.com/products/search"
      method: GET
      argsToUrlParam: true
    responseTemplate:
      body: |
        # Product Search Results

        Found {{.total}} products matching "{{.query}}", below are the {{len .products}} most relevant results:

        {{range $index, $product := .products}}
        ## {{add $index 1}}. {{$product.name}}

        **Price**: {{if $product.onSale}}~~{{$product.compareAtPrice}} {{$product.currency}}~~ **{{$product.price}} {{$product.currency}}** (Save {{percentage $product.compareAtPrice $product.price}}%){{else}}{{$product.price}} {{$product.currency}}{{end}}

        **Brand**: {{$product.brand.name}}

        **Stock Status**: {{if eq $product.availability "in_stock"}}In Stock{{else if eq $product.availability "low_stock"}}Low Stock{{else}}Out of Stock{{end}}

        {{if gt (len $product.ratings.reviews) 0}}**Rating**: {{$product.ratings.averageRating}}/5 ({{$product.ratings.reviewCount}} reviews){{end}}

        {{$product.description | truncate 200 "..."}}

        {{if gt (len $product.highlights) 0}}**Product Features**:
        {{range $highlight := $product.highlights}}
        - {{$highlight}}
        {{end}}{{end}}

        {{end}}

        {{if gt .total (len .products)}}
        There are more results not displayed; you can adjust the search criteria for more precise matches.
        {{end}}

Here’s a real response example for a smart watch product, showing how the tuned template handles the raw data:

# Product Search Results

Found 128 products matching "smartwatch." Below are the 10 most relevant results:

## 1. TechFit Pro Smartwatch

**Price**: ~~1299 CNY~~ **899 CNY** (Save 30.8%)

**Brand**: TechFit

**Stock Status**: In Stock

**Rating**: 4.7/5 (342 reviews)

The TechFit Pro smartwatch is equipped with a high-definition color touchscreen, supports heart rate monitoring, blood oxygen detection, various sports mode tracking, and sleep analysis. It has a waterproof design with a battery life of up to 7 days.

**Product Features**:
- High-definition AMOLED touchscreen
- 7-day ultra-long battery life
- Heart rate and blood oxygen monitoring
- 30 sports modes
- 5ATM waterproof

## 2. FitBit Versa 3

**Price**: 1499 CNY

**Brand**: FitBit

**Stock Status**: In Stock

**Rating**: 4.5/5 (287 reviews)

The FitBit Versa 3 smartwatch integrates GPS positioning, supports 24/7 heart rate monitoring, has a built-in voice assistant, can receive phone calls, and has a battery life of up to 6 days.

**Product Features**:
- Built-in GPS
- Voice assistant functionality
- Call answering
- 20+ sports modes
- Waterproof up to 50 meters

## 3. Apple Watch Series 7

**Price**: 2999 CNY

**Brand**: Apple

**Stock Status**: Low Stock

**Rating**: 4.9/5 (1243 reviews)

The Apple Watch Series 7 features a larger display, faster charging speed, IP6X dust resistance certification, swimming-level waterproofing, and all-day blood oxygen monitoring and electrocardiogram functions.

**Product Features**:
- Retina-level OLED display
- Fast charging
- Electrocardiogram and blood oxygen monitoring
- Fall detection and emergency SOS
- Supports Apple Pay

There are more results not displayed; you can adjust the search criteria for more precise matches

With this structured response format, LLM can clearly understand the key information of each product without being overwhelmed by a large amount of technical details and raw JSON structure.

Based on the above case, we can extract the following tuning strategies for MCP configuration:

🔹 Identify and extract core fields: Analyze the information that users truly need, removing technical details and internal data.

🔹 Transform technical jargon: Convert technical language and jargon into descriptions that LLM can easily understand.

🔹 Add contextual information: Help LLM understand the completeness and scope of the data.

🔹 Structure key information: Use hierarchical structures to make the importance and relationships of the information clear.

2.3 Enterprise-level Capabilities

Batch conversion of existing APIs and refined tuning solve production efficiency issues, while enterprise-level capabilities address diverse enterprise needs such as authentication, traffic control, protocol version upgrades, etc.

Higress is built on Envoy, inheriting the mature capabilities of traditional API gateways in authentication, authorization, rate limiting, and observability, while being optimized for AI scenarios. These capabilities closely align with the needs of MCP services:

• Unified authentication and authorization: Higress provides an OAuth2 plugin that meets the new MCP protocol's requirements for authentication and authorization. Developers don't need to write complex authentication code; simply configuring can achieve secure user authentication and access control.

• Refined traffic control: Through Higress’s rate limiting plugin, different invocation quotas can be set for different users and tools to prevent resource abuse and service overload.

• Full-link observability: Higress integrates open-source observability solutions like Prometheus and OTel, providing complete metrics monitoring and distributed tracing capabilities, allowing operations teams to monitor the health and performance of MCP services in real-time.

• High performance: Compared to traditional gateways, Higress has stronger performance, supporting horizontal scaling and can better handle high invocation volumes of MCP.

• Audit logs: Records all tool invocation behaviors to meet compliance requirements, while providing data support for security analysis.

Advantages of Protocol Offloading

There are currently two versions of the MCP protocol, which are version 20241105 and 20250326. In terms of protocol development, the speed of standard sedimentation and optimization is not keeping pace with the rapidly evolving MCP Server ecosystem. This poses a hidden danger for early adopters of MCP, as building a large number of MCP Servers means they will face a heavy upgrade and transformation workload in the future when protocol version upgrades are required.

As a gateway, Higress can offload MCP protocols, just like serving traditional API gateways, uniformly offloading client protocols of HTTP1/HTTP2/HTTP3 to shield multiple versions of protocol details.

Higress can support both 20241105 and 20250326 versions of the MCP protocol simultaneously at one access point. At the transport layer of the protocol, it supports POST+SSE mode and the latest Streamable HTTP mode, native support for WebSocket, and maintains connection after configuration changes, providing graceful disconnection, thus preparing technical reserves for the evolution of MCP protocols.

2.4 Hosted MCP Servers

When deploying MCP Servers, in addition to using compute platform products provided by cloud vendors, users can also choose to host on Higress, with significant advantages in deployment and operation:

• Elastic scaling: Based on Kubernetes, it has automatic scaling capabilities to adjust the number of instances according to traffic, ensuring high availability of services and efficient resource utilization.

• Gray-release: Supports gray releases and A/B testing of MCP Servers, reducing update risks.

• One-click deployment: Provides Helm Chart to simplify deployment processes and lower operation thresholds.

• High performance: Can easily handle tool invocation requests in the hundreds of thousands per second while maintaining sub-millisecond response times, meeting the performance needs of enterprise applications.

03 How to Use Higress MCP Marketplace

All services that have been released to the Higress MCP Marketplace are maintained by Higress, and use cases along with the number of free usages are provided (one account can access for free 10 times a day).

Taking the stock assistant as an example, this is a multi-functional API service specifically designed for the stock, futures, and foreign exchange markets. It offers various tools, including candlestick charts, quotes, rankings, etc., to help users obtain real-time and historical data, perform technical analysis, and support decision-making. Through these tools, users can easily access relevant information from the A-shares, Hong Kong stocks, US stocks, global indices, domestic and foreign futures, and foreign exchange markets, specific tools include:

A-share Candlestick

• Purpose: Provides candlestick data for A-shares across different time periods (such as 1-minute, 5-minute, daily candlesticks, etc.).

• Usage scenarios: Technical analysis, transaction strategy formulation, historical data backtesting, etc.

A-share Adjusted Candlestick

• Purpose: Provides candlestick data for A-shares after adjustment.

• Usage scenarios: Long-term investment analysis, fundamental analysis, etc.

A-share Quotes

• Purpose: Provides real-time quote information for A-shares.

• Usage scenarios: Real-time monitoring, rapid trading decisions, etc.

A-share Rankings

• Purpose: Provides rankings of A-shares based on specific conditions (such as price change rate, transaction volume, etc.).

• Usage scenarios: Discovering hot stocks, market trend analysis, etc.

After logging in to mcp.higress.ai, you can enter the page and directly copy the URL generated on the right. We also provide SSE access points and Streamable HTTP access points, supporting access based on client use conditions.

If the trial usage limit has been reached and you wish to continue using it, you can refer to the instructions in the overview tab to apply for an API Key, configure it below, and then regenerate the URL:

04 Future Plans for Higress MCP Marketplace

• Collaborate with multiple API providers such as Aliyun Cloud Market to continuously provide more free trial quotas for different Servers.

• Launch more commonly used Servers in coding scenarios, and deeply integrate with Tongyi Lingma to facilitate user connections.

• To be integrated with the Nacos MCP Registry, allowing enterprise users to quickly start MCP Servers with equivalent capabilities.

05 Q&A

Q1: What are the benefits of uploading my MCP Server to Higress MCP Marketplace?

A1: Gain traffic from Higress and theenterprise version of the Aliyun API Gateway, getting more invocation volume from around the world. At the same time, we will recommend it to the Aliyun MCP Market, such as the MCP market of Aliyun Cloud Market, reducing your communication costs.

Q2: Our MCP Server has already been listed in other MCP markets. What additional development work is needed to upload it to Higress MCP Marketplace?

A2: If you have existing API documentation, no additional development work is needed. Using the API to MCP tool, it can automatically convert to the YAML configuration of the MCP Server. Tool link:

https://github.com/higress-group/openapi-to-mcpserver

Q3: How can I upload the MCP Server developed based on Higress to Higress MCP Marketplace?

A3: You can refer to this example to submit a PR to our open-source repository:

https://github.com/alibaba/higress/pull/2080/files

Then contact our staff to discuss cooperation matters. Ding group number 107690002780, or WeChat (ID: zjjxg2018)

Q4: Which MCP Clients can invoke MCP Servers implemented based on Higress, such as Tongyi Lingma, Cline, Cherry Studio, etc.?

A4: As long as the client supports POST+SSE mode and Streamable HTTP mode, it can invoke.

Q5: Which AI application development platforms support the integration of MCP Servers implemented based on Higress, such as Bailian, Dify, Spring AI Alibaba, LangChain, etc.?

A5: As long as the development platform supports connecting to the MCP protocol, it can be integrated.

Q6: Which large models can recognize MCP Servers implemented based on Higress, such as Tongyi, DeepSeek, OpenAI, etc.?

A6: As long as the large model supports Function Call or can generate responses in a specified format (e.g., XML), it can recognize it.

Q7: Our MCP Server has been deployed. How can we use Higress's enterprise-level capabilities?

A7: Download Higress and refer to the official documentation for deployment; if your business is deployed on Aliyun, it is recommended to use Higress Enterprise Edition. For detailed comparison of differences: https://help.aliyun.com/zh/api-gateway/cloud-native-api-gateway/product-overview/product-comparison

Q8: Our MCP Server has not yet been deployed. Can it be hosted directly on Higress?

A8: Yes, users do not need to deploy their own servers. Based on Higress's high performance and horizontal scaling capabilities, it can easily host thousands of MCP Servers. If you have more stringent performance, stability, and security requirements, Higress Enterprise Edition is recommended.