Open source Remote MCP Server one-stop hosting is here!

The ecological development of the MCP (Model Context Protocol) protocol is advancing at an unprecedented speed. Recently, two key events have marked that MCP has moved from a de facto standard to an industry standard: on the one hand, OpenAI officially announced that it would follow Anthropic's MCP protocol, and on the other hand, Anthropic released a new version of the MCP protocol, which achieved significant improvements in the Remote MCP Server scenario.

OpenAI CEO Sam Altman confirmed on social media that OpenAI will integrate Anthropic's MCP protocol into its products (including the ChatGPT desktop application). Altman said: "MCP has received a good market response, and we are very happy to support this protocol in our own products. This protocol is now open in the Agents SDK, and support for the ChatGPT desktop application and Responses API will be launched soon!" This move means that developers can more easily build intelligent assistants that can call real-time data, such as enterprise-level chatbots or automated workflows.

At the same time, Anthropic released a new revision of the MCP protocol on March 26, 2025. The core highlight of this update is the Streamable HTTP transmission mechanism, which not only inherits the real-time message transmission capability of HTTP+SSE, but also makes the connection more stable and the data transmission more resilient through a series of optimizations. The new version of the protocol uses a single MCP endpoint to support both HTTP POST and GET requests, implements session management by forcing the use of the Mcp-Session-Id header, and supports batch requests, responses and notifications, as well as the recoverability of SSE streams.

The combined effect of these two events is accelerating the prosperity of the MCP ecosystem. According to statistics, as of February 2025, more than 1,000 community-built MCP servers are available. With the addition of OpenAI, the MCP protocol is expected to become a universal standard for connecting AI models with the outside world, just like USB, HTTP or ODBC in their respective fields.

MCP Server is changing the capabilities of AI applications, transforming AI from a simple conversational machine into an intelligent assistant that can understand user needs and help complete practical tasks. The following are typical application scenarios of MCP Server in different fields:

In an enterprise office environment, MCP Server can connect to various internal systems, such as mail servers, calendars, document management systems, etc., enabling AI assistants to:

• Meeting Management
  Automatically record meeting content, generate meeting minutes, and create to-do items based on discussions
• Mail processing
  Categorize important emails, draft replies, and set reminders
• Document collaboration
  Find information in team documents, suggest edits, and track changes

For example, a manager can ask the AI ​​assistant: "Collate the key points from all sales meetings last week and create a list of action items." The AI ​​assistant automatically completes this task by accessing the meeting notes system and project management tools through MCP Server.

In smart home and industrial IoT environments, MCP Server can connect various smart devices and sensors to achieve:

• Equipment status monitoring
  Track equipment operating conditions in real time and predict maintenance needs
• Cross-device collaboration
  Coordinate multiple devices to complete complex tasks
• Environmental Optimization
  Automatically adjust environmental parameters based on user preferences and usage patterns

Users can request: "When I get up at 7 o'clock tomorrow morning, turn on the coffee machine 20 minutes earlier and adjust the temperature in the living room to 22 degrees." The AI ​​assistant communicates with the smart home system through the MCP Server to schedule these tasks.

Software development teams can use MCP Server to connect code repositories, CI/CD pipelines, and project management tools to improve development efficiency:

• Code Assist
  Generate code based on project context and provide refactoring suggestions
• Automated testing
  Generate test cases, execute tests and analyze results
• project management
  Track task progress, allocate resources, and predict delivery times

Developers can ask: "Analyze our code base, find all unhandled exceptions, and provide fix suggestions." The AI ​​assistant accesses the code repository through MCP Server, performs static analysis, and generates a detailed report.

With the development of the MCP ecosystem, the evolution from Local MCP Server to Remote MCP Server has become a trend. In order to better understand this transition, we need to first clarify the concepts and differences between the two types of MCP Servers.

Local MCP Server refers to the MCP server running on the user's local device. In this mode, the MCP client (such as Claude Desktop or Cursor) interacts with the MCP server through local process communication (stdin/stdout), and the server connects to various APIs and services on the Internet. This architecture is simple and direct, suitable for individual developers, but it has some limitations.

Remote MCP Server is an MCP server deployed in the cloud and can be accessed by users through the Internet. In this mode, the MCP client can be a wider range of web applications or mobile applications, which communicate with the remote MCP server through the HTTP protocol. Remote MCP Server usually integrates enterprise-level functions such as authentication and authorization, state management, and database access, and can provide services for multiple users.

This shift addresses many of the limitations of Local MCP Server in enterprise environments.

Although Local MCP Server is easy to use, it faces many challenges in enterprise-level applications:

• Local environment dependencies
  It is dependent on the user's local environment. For example, it is necessary to install Python or Docker to run the MCP Server, which is not friendly to non-technical users.
• Security Risks
  It is impossible for an enterprise to configure sensitive database credentials, API keys or other critical access tokens to each employee's local environment. This not only violates the principle of least privilege, but also greatly increases the risk of credential leakage.
• Consistency issues
  When multiple users need to access the same enterprise resources, it is difficult to ensure consistency in configuration and permissions, which can easily lead to inconsistent data or confusion in permissions.
• Maintenance costs
  Deploying and maintaining MCP Server for each user device requires a lot of IT resources, and version updates, security patches, and configuration changes need to be performed separately on each device.

Remote MCP Server solves the above problems through centralized deployment and management:

• Expand usage scenarios
  Non-technical users can use MCP capabilities anytime and anywhere through the Internet, such as through web pages or mobile applications.
• Centralized security control
  Enterprises can implement strict access controls, encryption, and auditing mechanisms on remote servers to ensure the security of sensitive credentials.
• Unified permissions management
  Through a centralized authentication and authorization system, enterprises can precisely control each user's access rights to different resources.
• Simplify deployment and maintenance
  Only the central server needs to be maintained, which greatly reduces the operation and maintenance costs and complexity.

Although remote MCP Server has many advantages, its implementation also brings new challenges, especially in terms of authentication and authorization, service reliability, and observability. This is exactly where API gateways excel, and Higress, as an AI-native API gateway, provides a complete open source MCP Server hosting solution:

This solution has been updated to Anthropic's official MCP Server project documentation. For details, please click: https://github.com/modelcontextprotocol/servers?tab=readme-ov-file#for-servers

Higress's MCP Server hosting solution adopts a layered architecture design, providing AI Agents with complete service access capabilities. At the top layer, various AI Agents (such as Claude, Cursor, Cline, etc.) interact with Higress through the standard MCP protocol. Higress implements a complete security and control layer internally, including core functions such as MCP session persistence, OAuth2 authentication, audit logs, rate limiting, and MCP routing.

At the service level, Higress supports three MCP Server access methods:

1. Built-in MCP Server implemented in Higress through Wasm plugin, suitable for connecting to the existing SaaS API ecosystem to provide MCP capabilities
2. Directly forward to external services that already support the MCP protocol, suitable for connecting to existing external MCP capabilities
3. Dynamically discover external MCP Servers through service registration centers (such as Nacos), and dynamically update MCP Server tool definitions through the capabilities of the Nacos configuration center. This is suitable for enterprises to upgrade traditional business APIs to MCP capabilities.

This flexible architecture design allows enterprises to choose the most suitable deployment method according to their own needs. They can enjoy the one-stop hosting service provided by Higress while maintaining the independence of the existing MCP Server.

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

• Unified authentication and authorization
  Higress provides an OAuth2 plug-in that can meet the authentication and authorization requirements of the new MCP protocol. Developers do not need to write complex authentication code, and can achieve secure user authentication and permission control with simple configuration.
• Fine flow control
  Through Higress's current limiting plug-in, different call quotas can be set for different users and different tools to prevent resource abuse and service overload.
• Full-link observability
  Higress integrates open source observation solutions such as Prometheus and OTeI, providing complete indicator monitoring and distributed tracing capabilities, enabling the operation and maintenance team to monitor the health and performance of MCP services in real time.
• Audit log
  Record all tool calling behaviors to meet compliance requirements and provide data support for security analysis.

One of the technical highlights of Higress is its support for WebAssembly (Wasm) plug-ins, which provides MCP Server with powerful expansion capabilities:

Through the Wasm plugin mechanism, developers can:

• Quickly add a new server
  MCP Server is developed in multiple languages ​​(currently MCP Server SDK in Go language is provided first), compiled into Wasm and dynamically loaded into Higress without restarting the service.
• Security Isolation
  Each Wasm plugin runs in an independent sandbox, so even if the plugin crashes, it will not affect the stability of the gateway.
• Dynamic Updates
  Thanks to Envoy's friendly handling of long connections and the dynamic update mechanism of the Wasm plug-in, the MCP Server logic can be updated in real time without interrupting traffic.

Currently, MCP has two versions of the protocol, 20241105 and 20250326. From the perspective of protocol development, the speed of standardization and optimization is difficult to keep up with the rapidly developing MCP Server ecosystem. This also buries hidden dangers for early adopters of MCP. When a large number of MCP Servers are built, they will need to carry out heavy upgrade and transformation work in the face of future protocol version upgrades.

The Higress gateway can help shield the details of multiple versions of the protocol. It can even implement the ability to convert MCP to REST/Dubbo and other protocols. All versions of the MCP protocol can be unloaded at the Higress gateway. Just like when we use Higress as a traditional API gateway, we can unload the client protocols of HTTP1/HTTP2/HTTP3.

Currently, Higress can support both MCP version 20241105 and version 20250326 on one access point. At the transport layer of the protocol, it supports the POST+SSE mode currently used by most AI tools, as well as the latest Streamable HTTP mode. We expect that in the future, as the application ecosystem requires real-time performance, full-duplex, two-way real-time communication protocols such as WebSocket will become more popular and widely adopted. Higress natively supports WebSocket, which can achieve full-duplex, two-way real-time communication, and Higress supports connection maintenance and graceful disconnection after WebSocket configuration modification. At that time, Higress's advantage in MCP session maintenance will become more prominent.

Higress's MCP Server Hosting solution has significant advantages in deployment and operation and maintenance:

• Elastic scaling
  Based on the automatic scaling capability of Kubernetes, the number of instances is automatically adjusted according to the traffic to ensure high availability of services and efficient resource utilization.
• Grayscale release
  Supports grayscale release and A/B testing of MCP Server to reduce update risks.
• One-click deployment
  Provide Helm Chart to simplify the deployment process and lower the threshold for operation and maintenance.
• MCP Server hosted by Higress can easily handle hundreds of thousands of tool call requests per second while maintaining millisecond response time to meet the performance requirements of enterprise-level applications.

The Higress community is actively promoting the development of the MCP ecosystem and plans to provide an open MCP market on higress.ai in mid-April 2025 to facilitate users to experience the advantages of Remote MCP Server. For example , this video directly enters an MCP Server access point deployed in Higress on the Cursor to connect to the database schema and write CRUD code:

Through the open MCP market, we can allow users to experience this capability without having to deploy Higress and MCP Server themselves.

Higress’ MCP Marketplace will provide:

• Rich MCP Server Ecosystem
  Covers common application scenarios, such as document processing, data analysis, code assistance, etc.
• One-click use
  Through the web interface, users can obtain the MCP Server access point with one click and use it in their own AI tools.
• Community sharing mechanism
  Developers can put their own MCP tools on the market after they have been reviewed by the open source community for users to use.

In addition to the open source community version, Alibaba Cloud API Gateway will soon be available on the MCP market to provide more comprehensive services for enterprise users:

• Enterprise-level SLA guarantee
  Provide 99.99% service availability commitment to meet the needs of key business of enterprises.
• Compliance Certification
  It has passed multiple security and compliance certifications to meet the strict requirements of industries such as finance and healthcare.
• Expert Support
  Provide 7x24 hours technical support and consulting services to help companies solve problems quickly.
• This allows enterprises to enjoy the SaaS MCP ecosystem while also building their own internal MCP Server ecosystem.

If you are a company that provides SaaS services and are looking to build an MCP Server so that users can pay to use your API in AI scenarios, the Higress community welcomes cooperation with you:

• Technical Support
  Provide technical guidance for MCP Server development, or let our open source community help you build it.
• Joint Promotion
  Promote partners’ MCP services on Higress official channels.
• Market entry
  Help your MCP Server to be listed and hosted in Higress's official MCP market.

Through open source , the Higress community is committed to promoting the prosperity and development of the MCP ecosystem, allowing more developers and companies to easily build and deploy MCP Servers and unleash the full potential of AI assistants.

With the promotion of industry giants such as OpenAI and Anthropic, and the support of open source projects such as Higress, the MCP protocol is moving from a de facto standard to an industry standard. In the future, we have reason to believe that MCP will become a key bridge connecting AI and the real world, bringing unprecedented intelligent experience to all walks of life.