a16z Insight: MCP reshapes the "universal interface" for next-generation AI applications

The “Tower of Babel” dilemma of AI tools: the urgent need for unified standards

The progress of artificial intelligence, especially large language models (LLM), has laid the foundation for building powerful AI agents. However, in the actual application of AI agents, the problem of tool fragmentation has become increasingly prominent. Developers customize business logic for each tool and system, and the integration cost is high, which seriously hinders the expansion of AI capabilities and the improvement of application efficiency.

As mentioned earlier: " As the underlying models become smarter, the agent's ability to interact with external tools, data, and APIs becomes increasingly fragmented: developers need to implement specific business logic for each system the agent integrates with. " This points out the core challenge facing the current AI tool ecosystem: the lack of unified interaction standards.

The prosperity of the Internet has benefited from the standardization of APIs, which have become the "common language" for interconnection between different software systems. However, the field of AI models has long lacked similar unified standards, making it difficult for AI agents to effectively use various tools and services. Breaking the "Tower of Babel" dilemma of AI tools and building a unified interaction protocol have become key issues that need to be urgently addressed in the field of AI.

MCP: Building a “Universal Language” for AI Agents

The Model Context Protocol (MCP) was born to provide an open, universal protocol that allows systems to provide context to AI models in a standardized way, enabling cross-integrated tool calls, data acquisition, and service interactions.  The goal of MCP is to create a "common language" for AI agents, lower the threshold for tool integration and use, and promote the interconnection of the AI ​​ecosystem.

The core mechanism of MCP: Standardized protocol is the key

The core of MCP is to  define a set of standardized communication protocols , which regulate the interaction between AI models and external tools and services. By following the MCP protocol, different tools and services can be connected to AI agents in a unified way, realizing plug-and-play functional expansion.

Resend MCP Server working principle example: MCP client interacts with Resend service through Server

Resend MCP Server acts as an intermediary layer, connecting the MCP client and the Resend mail service. The client sends a standardized request to the Server via the MCP protocol, and the Server is responsible for protocol conversion and API calls, and finally returns the result to the client. This model simplifies the integration of the client and the Resend service and reduces the complexity of development.

Inspiration from LSP Protocol: Learning from Successful Experiences

The design of MCP is inspired by the Language Server Protocol (LSP). The successful application of LSP in the field of code editors provides valuable experience for the development of MCP.

Application of LSP protocol in code editors: Standardized protocol promotes the prosperity of code editor ecosystem

The LSP protocol defines the communication standard between the code editor (client) and the language server (server), realizes the standardized integration of functions such as code completion and syntax checking, and greatly promotes the prosperity of the code editor ecosystem. MCP draws on the ideas of LSP and hopes to establish similar standardized protocols in the field of AI tool interaction and build an open and interconnected AI tool ecosystem.

Agent-centric and autonomous execution: Beyond traditional API interaction

What makes MCP superior to LSP is its  agent-centric  and  autonomous execution  design concept. LSP is mainly a passive response protocol, while MCP is designed to support autonomous AI workflows.

MCP's Agent-centric Execution Model: Agents make autonomous decisions and flexibly orchestrate tools

Based on the context, AI Agent can autonomously select tools, plan execution paths, and complete complex tasks. MCP also introduces  a human-in-the-loop  mechanism, allowing human intervention to improve the flexibility and security of the system. This agent-centric design that supports autonomous execution and human-machine collaboration is the key feature that distinguishes MCP from traditional API interactions.

Application scenarios of MCP: empowering developers and innovating user experience

The MCP protocol has a wide range of application scenarios and shows great potential in optimizing developer workflows and innovating user experience.

Developer Center Workflow: A Tool for Improving Efficiency

MCP Server provides developers with a convenient way to integrate various tools in the IDE, meeting the developer's need to " complete X without leaving the IDE ".

Cursor Agent uses Browsetools for debugging: improving code debugging efficiency

For example, developers can use  Postgres MCP server  to execute SQL queries in IDE, use  Upstash MCP server  to manage cache, and use  Browsertools MCP  to debug code. In addition, tools such as Web crawling MCP server  and  document automatic generation MCP server  can help developers quickly add context information to coding agents and improve the quality and efficiency of code generation. MCP Server lowers the threshold for tool integration and improves developer work efficiency.

Innovative user experience: “Everything App” becomes possible

MCP not only serves developers, but also brings a new experience to non-technical users.  The emergence of clients such as Claude Desktop has lowered the threshold for using MCP.

Highlight client integration Notion MCP example: content pipeline innovation UX pattern

The @ command of the Highlight client   demonstrates a new UX mode. Users can call MCP Server through the @ command and transmit the generated content to other applications in a pipeline manner, realizing innovation in the content creation process. The application of Blender MCP server  demonstrates the potential of MCP in lowering the threshold for using professional tools. Users can drive Blender for 3D modeling through natural language commands. In the future, MCP is expected to spawn more innovative applications in the fields of customer support, marketing copywriting, design, image editing, etc.

MCP ecosystem market landscape: developer tools first, huge potential for commercial applications

At present, MCP clients are mainly concentrated in the field of developer tools, and servers are mainly deployed locally. With the maturity of the protocol and the improvement of infrastructure, the application scenarios of MCP will be more extensive.

MCP Ecosystem Scan: Development Status and Market Opportunities

The MCP ecosystem has begun to take shape, with development momentum in terms of clients, services, infrastructure and markets.

Client: Developer tools first, commercial applications ready to go

At present, high-quality MCP clients are mainly concentrated in  the field of coding-centric  developer tools, such as IDEs such as Cursor. In the future, with the popularization of the MCP protocol and the improvement of user awareness, MCP clients for  commercial applications  are expected to usher in rapid development, such as professional clients for customer support, marketing, design and other fields.

Server side: Local first, remote is the future trend

Currently, MCP Server is mainly  deployed locally and serves a single user  . This is because the MCP protocol currently mainly supports SSE and Command-based connections. With the support of the MCP protocol for  Streamable HTTP transport  and  the maturity of remote hosting  solutions, remote MCP Server is expected to become a future development trend, reducing deployment and operation and maintenance costs and improving the scalability and usability of the server.

Infrastructure and Market: The Cornerstone of Ecological Prosperity

The prosperity of the MCP ecosystem is inseparable from perfect  infrastructure and market  support.  Platforms such as mcpt ,  Smithery  and  OpenTools  launched by  Mintlify are building the MCP Server market to solve the problem of server discovery and distribution. Tools such as Mintlify ,  Stainless  and  Speakeasy are committed to lowering the threshold for generating  MCP Servers  . Platforms such as Toolbase  focus on  connection management and simplify local MCP key management and agents. The improvement of these infrastructures and markets will provide strong support for the rapid development of the MCP ecosystem.

The future of MCP: challenges and prospects

Although MCP shows great potential, its development still faces many challenges and requires continuous improvement of the protocol and ecosystem.

Hosting and multi-tenancy: a key step towards enterprise-level applications

MCP needs to better support  multi-tenant architecture to meet the needs of enterprise-level applications.  Remote server hosting  is a short-term solution to improve usability, but enterprise users   have higher requirements  for data security and compliance , and need to support enterprise self-hosted  deployment mode. Building  an at-scale server deployment and maintenance tool chain  is the key to promoting the popularization of MCP in enterprise-level applications.

Identity authentication: building a secure and reliable foundation

The MCP protocol needs to define  a standardized authentication mechanism to solve the identity authentication problem between the client and the server, and between the server and the third-party API.  Standard protocols such as OAuth 2.1  can serve as the basis for MCP identity authentication. A unified  Client/Tool/Multi-user authentication scheme  is essential to improving the security of MCP.

Authorization management: achieving more refined permission control

MCP currently lacks a built-in  permission model , and the access control granularity is relatively coarse. In the future, it is necessary to introduce  fine-grained permission control mechanisms , such as role-based access control (RBAC) or policy-based access control (PBAC), to meet the permission management needs of complex application scenarios.

Unified Gateway: Improving Scalability and Security of the Ecosystem

With the popularity of MCP applications, the need for a unified gateway  is becoming increasingly urgent. The MCP gateway can centrally manage authentication, authorization, traffic management, and tool selection, improving  the scalability, security  , and  manageability of the system , which is especially important for multi-tenant environments.

Server discovery and ease of use: lowering the threshold for developers

Server discovery and ease of use  are key factors that restrict the popularity of MCP.  The introduction of the MCP Server registry and discovery protocol  is expected to solve the server discovery problem and lower the threshold for using MCP. Building  an MCP Server market  is also an important measure to improve the usability of servers.

Execution environment: Supporting complex AI workflows

The MCP protocol needs to support  the management of complex workflows  . Introducing  the concept of workflows  or developing an independent  MCP workflow engine can better support multi-step tool calls and enhance the AI ​​Agent's ability to handle complex tasks.

Standardized client experience: Improving user experience consistency

Standardized client experience  is crucial to the popularity of MCP. Defining  MCP client tool selection standards  and  UI/UX specifications and developing  a UI component library can improve the consistency and predictability of user experience and reduce development costs.

Debugging tools: improving MCP development efficiency

The lack of debugging tools  is one of the challenges faced by MCP Server development. Enhancing  client tracking capabilities and developing  MCP Server debuggers  and  remote debugging tools can improve the development efficiency and quality of MCP Server.

The far-reaching impact of MCP: Reshaping the future of AI tools

The popularity of MCP will have a profound impact on the construction, consumption, and business models of AI tools.

Developer Competitive Advantage: Transformation from API Design to Agent Tool Ecosystem

In the future, the competitive advantage of developer-first companies will   shift  from API design to the construction of agent tool ecosystems  . Providing high-quality, easy-to-find, and differentiated tools will become the key to developer competition.

Pricing model: market-driven dynamic instrument selection mechanism

MCP is expected to give rise to  a market-driven dynamic tool selection and pricing model . Agents can dynamically select tools based on factors such as speed, cost, and relevance, promoting the optimization and modular development of the tool market.

Documentation: Becoming the new core of AI infrastructure

Documentation  will become a core component of the MCP infrastructure.  Machine-readable documentation  (such as llms.txt) will become the key for agents to understand and use the tool, and the quality of documentation will directly affect the competitiveness of the tool.

API and Tooling: The Leap of AI Abstraction Level

Tooling  will become a more important abstraction level in the Agent era, and API will retreat to the background and become the basic support for Tooling. The design of MCP Server will be more  scenario-oriented and use case-centric , rather than just a simple encapsulation of API.

Hosting Model: A New Paradigm for Agent Workflow Optimization

A new hosting model for agent workflows   will emerge, which needs to support multi-step workflows, recoverability, retries, and long-running task management, and achieve real-time load balancing across MCP servers to optimize cost, latency, and performance.

Seize the opportunity of AI tools as a “universal language” and jointly draw a new blueprint for intelligent applications

The MCP protocol is reshaping the AI ​​Agent ecosystem and is expected to become the default interface for AI-to-Tool interactions, unlocking a new generation of autonomous, multimodal, and deeply integrated AI experiences. 2025 will be a critical year for the development of MCP, and we look forward to seeing   breakthroughs in the MCP ecosystem in terms of market, certification, and execution standardization .

We sincerely invite developers, enterprises and researchers to pay attention to and participate in the construction of the MCP ecosystem, and jointly welcome the arrival of the era of "universal language" of AI tools. What do you think about the future development of MCP? Feel free to share your views and expectations in the comments section.

Recommended Reading

• • A Deep Dive Into MCP and the Future of AI Tooling: https://a16z.com/a-deep-dive-into-mcp-and-the-future-of-ai-tooling/
• • Model Context Protocol Specification: https://spec.modelcontextprotocol.io/specification/2024-11-05/