MCP: The Protocol Expanding AI's Abilities

How Anthropic's open standard is making AI systems more powerful through seamless integration

Agents are all the rage. The industry continues to build infrastructure for these increasingly capable systems to collaborate, reason, and execute complex workflows. Recently, Amazon Web Services announced the general availability of multi-agent collaboration on Amazon Bedrock, exemplifying this trend.

AWS describes their solution as enabling developers to "build, deploy, and manage networks of AI agents that work together to execute complex, multi-step workflows efficiently." Their system allows "specialized agents that communicate and coordinate under the guidance of a supervisor agent," breaking down complex processes into manageable sub-tasks that can be processed in parallel.

This focus on agentic AI extends throughout AWS's organization. Recently, AWS formed a new group focused specifically on developing agentic artificial intelligence to enhance automation for users and customers. This initiative aims to enable systems to perform tasks independently without user prompts.

As companies like AWS focus on agent collaboration frameworks and dedicated organizational structures, a practical standard is emerging in the background that helps AI applications interoperate with the systems around them.

On November 25, 2024, Anthropic introduced the world to the Model Context Protocol, an open standard designed to connect AI assistants directly to various data sources, such as content repositories, business tools, and development environments.

What is MCP?

The Model Context Protocol acts as a universal connector between AI models and external data sources or tools. It provides a standardized way for language models to access and interact with various external systems, enabling plugin and extension capabilities for AI applications.

Before MCP, advanced AI models were isolated by default. Each new data source or tool required custom integration work. MCP replaces this fragmented approach with a single, standardized protocol that enables two-way, real-time connections between AI systems and the digital world they need to access.

While currently showcased in Claude Desktop chat, MCP is not tied to any specific AI system. Application developers can implement MCP support to allow for customized integration into a larger ecosystem. The protocol works equally well for client-side applications and server-side solutions, providing flexibility for various implementation approaches.

Anthropic's release includes several key components that make MCP implementation accessible:

• A formal specification and SDKs for developers to implement MCP in multiple programming languages

• Local MCP server support integrated directly into Claude Desktop apps

• An open-source repository of pre-built MCP servers for popular systems

In the three months since its announcement, MCP has gained significant momentum with early adopters integrating it into their systems and a growing community of developers contributing new servers and implementations.

With MCP, an AI assistant can:

• Read and analyze documents from your local files

• Search code repositories

• Query databases

• Interact with communication tools like Slack

• Access web content

• Control browser automation

All through a consistent, secure interface without requiring model retraining. For application developers, MCP offers a pathway to create extensible AI applications that can connect to a growing ecosystem of tools and services.

Perplexity Adopts MCP, Signaling Broader Industry Movement

The adoption of MCP received a boost on March 12, 2025, when Perplexity AI announced their implementation of the protocol for their Sonar API. This move signals MCP's growing acceptance beyond Anthropic's Claude ecosystem.

Perplexity stated simply: "We've built an MCP server for Sonar, giving AI assistants real-time, web search research capabilities. Powered by Perplexity, Claude can now search the web and deliver real-time and accurate insights on demand."

This implementation represents a shift in the industry toward standardized AI interactions and the growing consensus around MCP as a unifying protocol. With Perplexity's MCP server, Claude and potentially other MCP-compatible assistants can now leverage Perplexity's search capabilities, demonstrating the practical benefits of the protocol's interoperability.

As an aside, Perplexity has increasingly adopted an expand-and-integrate-everywhere approach to their offering. They recently announced a phone partnership with Telekom for an AI Phone integrated with Perplexity Assistant, launched a Windows app, and have teased a Perplexity browser coming soon. Their MCP implementation aligns with this broader strategy of expanding their AI service through numerous integrations, similar to how Netflix and YouTube expanded by being pre-installed on every device. This expansion strategy deserves its own analysis another day, but it contextualizes why Perplexity would embrace an open standard like MCP.

The Burgeoning Ecosystem of MCP Servers

MCP features a rapidly expanding ecosystem of servers connecting AI models to various tools and data sources. These servers extend AI capabilities without requiring changes to the underlying models.

The MCP ecosystem now encompasses servers across multiple categories:

• File Systems: Secure interactions with local and cloud storage (Google Drive, Dropbox, OneDrive)

• Version Control: Integration with repositories (GitHub, GitLab, Bitbucket)

• Databases: Interaction with various data stores (PostgreSQL, MongoDB, MySQL)

• Communication: Messaging capabilities via popular platforms (Slack, Discord, Teams)

• Search & Web: Access to online information (Brave Search, Google Search, Wikipedia)

• Monitoring: Performance tracking and error analysis (Sentry, Datadog, New Relic)

• Location Services: Geolocation and mapping (Google Maps, OpenStreetMap)

• AI Services: Integration with additional AI capabilities (OpenAI, EverArt, DeepSeek)

This proliferation has occurred quickly. Anthropic open-sourced dozens of pre-built MCP server connectors for popular systems, and the developer community has expanded this list. Early adopters like Block and Apollo have integrated MCP to connect AI with their internal systems, while developer tools companies (Zed, Replit, Codeium, Sourcegraph) are leveraging MCP to enhance AI agents' contextual awareness in coding tasks.

The Evolution to AI-First Architecture

MCP represents a fundamental shift in software architecture with AI. We're witnessing the emergence of an AI-first service-oriented architecture, where the model becomes both the interface and the orchestration layer.

Traditional SOA relied on WSDL and service discovery to enable system-to-system integration. MCP places AI at the center, allowing models to discover, comprehend, and utilize services through a standard protocol. The language model becomes the reasoning system that pulls context and executes actions through available servers.

This could change how we build AI-enabled systems. The AI becomes the primary interface, with traditional applications becoming services that the AI orchestrates.

Anthropic has taken the concept of function calling (popular with developers using OpenAI's APIs) and evolved it into a comprehensive protocol. By making MCP open and standardized, they've created an ecosystem of new capabilities and bridges to servers in a matter of months.

Automation and integration is moving to models, making them both the UI and the orchestration layer for complex workflows. The ability to combine multiple MCP servers creates powerful possibilities. An AI that can simultaneously access your filesystem, query a database, and communicate with your team through Slack can solve problems in ways that isolated systems cannot.

The OAuth Moment for AI Integration

The emergence of MCP resembles how OAuth transformed authentication. Before OAuth (and OIDC), the world of federated identity, SSO, and delegated scoped access grants was fragmented between enterprise-focused, committee-driven standards and numerous proprietary approaches. OAuth succeeded because it offered something practical and necessary while remaining simple to implement.

MCP appears poised for a similar trajectory. While committee-driven enterprise standards might offer theoretical completeness, MCP provides a practical solution to an immediate need: connecting AI models to the digital world in a standardized way.

OAuth succeeded because developers could actually use it. Similarly, MCP's straightforward JSON-based interface and clear conceptual model makes it accessible to many developers. The protocol aims to be good enough for most cases and better than the alternatives.

The pattern is familiar. A practical, open standard emerges to solve a pressing problem, gains adoption through ease of implementation, and eventually becomes ubiquitous through network effects.

Why You Should Pay Attention to MCP

If you're building with AI, integrating AI into existing systems, or trying to understand where the technology is headed, MCP demands your attention for several reasons:

1. Interoperability: MCP enables any supporting AI model to connect to any MCP-compliant tool or data source, simplifying the expansion of integration in AI applications.

2. Runtime extensibility: MCP allows tools to be added at runtime, even by end-users. This creates a more dynamic and flexible AI ecosystem, expanding the accessibility of function calling and incorporating additional context.

3. Separation of concerns: MCP promotes a cleaner architecture where models, tools, and data connectors are distinct pieces that can be updated or swapped independently.

4. User control and security: MCP includes mechanisms for user approval of tool actions, particularly important when dealing with sensitive data or systems.

5. Community momentum: The growing ecosystem of MCP servers and industry adoption suggests the protocol has reached a tipping point toward becoming a standard.

The Model Context Protocol serves as connective tissue that will enable AI to become increasingly useful by bridging the gap between models and the systems where our data and functionality reside.

While tangential to the agent platforms being developed by major cloud providers, startups, and open source frameworks, MCP offers a complementary solution that these agentic systems can leverage. AI agents can both consume MCP servers to access external systems and expose themselves as MCP servers to other agents. This creates a powerful ecosystem where specialized agents can offer their capabilities through a standardized protocol, allowing other agents to discover and incorporate these abilities into their workflows. As the agent ecosystem grows, MCP provides a practical foundation for interoperability that enables agents to build on each other's strengths without custom integration work.

There is a good chance MCP becomes the predominant way to expose custom functionality, context, and actions to AI-based applications. At the very least, it will be a blueprint and launching point for the evolution of a standard in this space. The advancement in AI includes how models connect to and interact with our digital world. The Model Context Protocol is another important step forward in the architecture of future AI-based applications and services.