AI Agents are Tooling Up for the Future with Anthropic's new Model Context Protocol

🔗 Originally published on LinkedIn

ChatGPT, Claude, and Gemini have captured the imagination of users and developers alike. But what began as powerful conversational tools are rapidly evolving into something far more capable: intelligent agents that can leverage other software to perform tasks and extract contextual insights.

Claude, in particular, has demonstrated these capabilities by going the extra mile and introducing the Model Context Protocol (MCP), which has been open-sourced. Hot on the heals of the recently released Claude Computer Use, MCP is designed to help AI agents interact with different contexts more seamlessly, enabling them to make informed decisions and take actions based on a broader understanding of their environment.

Anthropic's vision for MCP is to create a standard that allows agents like Claude to integrate context from multiple sources, making them more versatile and capable in real-world applications.

This transition signals a potential paradigm shift in how we interact with digital systems, one that aligns closely with the concept of agentic computing—a model of software design and interaction that emphasises autonomy, goal-orientation, and adaptability.

The Current Landscape: Tools Enhanced by LLMs

Today, many existing tools integrate large language models (LLMs) to enhance functionality. For example:

• Microsoft Copilot: Embedded into Office applications, Copilot assists with tasks like drafting emails, creating presentations, and analysing spreadsheets.
• Notion AI: Seamlessly generates, organises, and edits content within the productivity platform.
• Adobe Firefly: Augments creative tools by enabling users to generate assets through natural language prompts.

These integrations demonstrate how traditional software can leverage LLMs to provide enhanced experiences. The LLM serves as an auxiliary engine, empowering users with generative and reasoning capabilities. However, these tools are bound by the frameworks of their parent applications. The AI’s role is secondary—it extends the tool but does not redefine the nature of interaction.

The Emergence of LLM Agents: Flipping the Paradigm

Where traditional tools enhance their capabilities by integrating LLMs, applications like ChatGPT and Claude are starting to represent the inverse: LLM agents that make use of external tools. This inversion is subtle but significant. Instead of being bound within a single application, these agents function as general-purpose interfaces capable of:

1. Executing Tasks Across Platforms: For example, connecting with APIs to send emails, schedule meetings, or pull financial data from external systems.
2. Reasoning with Context: By interacting with various tools, they gather domain-specific knowledge or contextual data to refine their reasoning and responses.
3. Dynamic Adaptation: Unlike traditional software, which relies on pre-programmed workflows, LLM agents can adapt their approach to meet novel or unexpected challenges.

An LLM agent like ChatGPT with plugins or Claude with MCP integrations embodies this shift. These agents are no longer merely responding to queries within a confined dataset. Instead, they orchestrate interactions across tools, functioning as both decision-makers and task executors.

It exemplifies how agentic computing is transforming AI from conversational assistants into proactive, goal-oriented actors.

Linking to Agentic Computing

Agentic computing provides a conceptual framework for understanding this evolution. At its core, agentic computing emphasises autonomous, goal-directed entities that can act on their environments. These agents:

• Perceive Context: They gather information from their environment (in this case, external tools or applications) to understand the task at hand.
• Plan Actions: Based on their goals and context, they decide on the optimal sequence of actions to achieve their objectives.
• Execute Autonomously: They carry out tasks without constant user intervention, adapting to changes dynamically.

This is where the juxtaposition becomes clear: traditional tools incorporating LLMs provide a degree of intelligence within predefined boundaries. In contrast, LLM agents embody the agentic model by stepping outside those boundaries, leveraging other tools as extensions of their capabilities. This shift moves us closer to a world where applications are no longer siloed but interconnected through AI agents acting as central orchestrators.

👉 What is Agentic Computing?

Challenges and Opportunities

The rise of LLM agents brings both promise and complexity. Key challenges include:

• Context Overload: How do agents prioritise and synthesise information from multiple sources without becoming overwhelmed or inefficient?
• Reliability: Trusting agents to execute tasks autonomously requires a higher level of accuracy and error management than today’s tools provide.
• Integration Complexity: Developing universal interfaces between agents and tools is no small feat, requiring standardisation and collaboration across platforms.

However, the opportunities are equally profound. Imagine an agentic future where:

• A legal AI agent dynamically references case law from multiple jurisdictions and drafts contracts tailored to a client’s specific needs.
• A healthcare agent pulls patient data from various systems, synthesises insights, and recommends personalised treatment plans.
• A marketing agent coordinates across analytics platforms, ad tools, and customer relationship management (CRM) systems to execute campaigns end-to-end.

These scenarios demonstrate how agents could reshape industries by enabling seamless, adaptive interactions across fragmented digital ecosystems.

Real-World Example: Claude Using MCP to Order Groceries

A recent development involving Claude and the Model Context Protocol (MCP) is an excellent illustration of this agentic evolution in action. In this case, Claude was set up to autonomously order groceries from Amazon Fresh by using MCP to connect to the appropriate service. The code for this Amazon Fresh MCP server can be found here. This example underscores how MCP can serve as a bridge, allowing an LLM to move beyond static responses and perform tangible, user-requested actions in the real world.

This is a practical manifestation of agents as commanding interfaces that dynamically engage with their environment. By using MCP, Claude could perceive user intent, gather context (like the user's grocery list), and autonomously interact with an external service to fulfil the request—all without the need for constant user intervention. It exemplifies how agentic computing is transforming AI from conversational assistants into proactive, goal-oriented actors.

Microsoft has already called Copilot 'The UI for AI,' which raises an interesting question: Could these mega agents like ChatGPT, Claude, and Gemini become the equivalent of the browser or operating system in an agentic world?

Direction of Travel: What Comes Next?

The future likely holds a convergence of the two paradigms. Tools will increasingly integrate LLMs to provide sophisticated capabilities, while LLM agents will continue to expand their reach by interacting with an ever-broader range of tools. Over time, these trends may lead to:

1. Standardised Protocols for Agent-Tool Interaction: A universal API layer could emerge, enabling agents to interact with tools more seamlessly.
2. Hybrid Models: Software could function both as standalone tools and as agent-integrated modules, offering flexibility depending on user needs.
3. Greater Autonomy: As trust and reliability improve, agents will take on more complex, high-stakes tasks with minimal user intervention.

Ultimately, the line between applications and agents may blur, giving rise to a new era of computing defined by fluid, interconnected systems orchestrated by intelligent agents.

Conclusion

As applications like ChatGPT, Claude, and Gemini continue to evolve, their trajectory points toward a future where AI agents are the primary interface for human-computer interaction. Microsoft has already called Copilot 'The UI for AI,' which raises an interesting question: Could these mega agents like ChatGPT, Claude, and Gemini become the equivalent of the browser or operating system in an agentic world? In other words, could they become the main entry point for doing everything, orchestrating interactions across the digital ecosystem?

By leveraging other applications as tools and sources of context, these agents embody the principles of agentic computing, challenging the traditional boundaries of software design. The juxtaposition between tools enhanced by LLMs and LLM agents leveraging tools highlights a shift in focus: from enhancing individual capabilities to orchestrating systems. This shift, while still in its infancy, has the potential to redefine the way we interact with technology, making it more intuitive, adaptive, and capable.

In the years to come, the direction of travel will depend on how we balance innovation with reliability, autonomy with control, and intelligence with trust. But one thing is certain: the era of AI agents is not a passing trend—it is the next chapter in the evolution of computing.

This article  was originally written and published on LinkedIN by Kevin Smith, CTO and founder of Dootrix.