Can’t stop hearing about AI Agents and MCP? there’s a good reason for it. We’re witnessing a shift towards autonomous systems capable of handling increasingly complex tasks. However, a critical challenge has emerged: how do these intelligent agents interact with the vast landscape of existing tools, data, and even each other?
Two recent developments are poised to revolutionise this: Anthropic’s Model Context Protocol (MCP) and Google’s Agent2Agent (A2A) protocol. For anyone watching the future of AI tooling – especially from a venture perspective – it’s important to understand these developments and how they can impact companies.
The Fragmentation Problem: each tool requires bespoke solutions
As foundational AI models become more intelligent, their ability to leverage external resources becomes crucial. However, today’s reality is fragmented. Developers are forced to build bespoke integrations for each tool an agent needs to use, creating a complex web of connections. This “NxM problem,” as eloquently described, where N represents AI models and M stands for tools, leads to redundant development, excessive maintenance, and inconsistent implementations. Just as the internet needed APIs to create a shared language for software, the AI ecosystem desperately needs standardisation.
Enter the Model Context Protocol (MCP): A Universal Remote for AI
Introduced in November 2024, MCP is an open protocol that provides a generalisable way for systems to offer context to AI models. Think of it as a standardised interface that defines how an AI model can call external tools, fetch data, and interact with services.
MCP is an open protocol that allows systems to provide context to AI models in a manner that’s generalizable across integrations
Inspired by the success of the Language Server Protocol (LSP) in the coding world, MCP extends beyond reactive responses to support autonomous AI workflows. This means AI agents can intelligently decide which tools to use, in what order, and how to chain them together to achieve a goal. Crucially, MCP also incorporates human-in-the-loop capabilities for added control and data input.
The implications are profound. With the right set of MCP servers, any MCP client can become an “everything app“. Imagine a code editor like Cursor seamlessly transforming into a Slack client, an email sender, or an image generator simply by connecting to the relevant MCP servers. Developers can stay within their IDEs to manage databases (Postgres MCP server), caching (Upstash MCP server), or even debug code with live environment access (Browsertools MCP). Beyond developer tools, MCP opens doors for net-new experiences for non-technical users, with clients like Claude Desktop making MCP-powered tools more accessible. We’re already seeing use cases in generating UI, creating hero images, and even enabling natural language interaction with complex software like Blender.
The MCP ecosystem is rapidly evolving, with a growing number of clients (primarily coding-centric for now) and servers being developed. Marketplaces like Mintlify’s mcpt, Smithery, and OpenTools are emerging to facilitate the discovery and sharing of MCP servers, much like package managers for software. Infrastructure and tooling providers are also stepping up to address scalability, reliability, and accessibility.
However, MCP is still in its early stages. Challenges around hosting, multi-tenancy, authentication, authorisation, server discoverability, execution environment, and debugging still need to be addressed. The next iteration of the protocol promises to tackle many of these foundational issues.
Google’s Agent2Agent (A2A): Agents Talking to Agents
Building on the momentum of agentic AI, Google announced yesterday their Agent2Agent (A2A) protocol, an open interoperability protocol designed for seamless collaboration between AI agents across diverse frameworks and vendors. Aimed at enterprises grappling with siloed AI systems, A2A seeks to standardise communication, automating complex workflows and boosting productivity. With support from over 50 technology partners, including major players like Salesforce, SAP, and ServiceNow, A2A provides a universal framework for agents to securely exchange information and coordinate actions.
From their official Google Cloud Next announcement:
We’re proud to be the first hyperscaler to create an open Agent2Agent (A2A) protocol to help enterprises support multi-agent ecosystems, so agents can communicate with each other, regardless of the underlying technology. More than 50 partners including Accenture, Box, Deloitte, Salesforce, SAP, ServiceNow, and TCS, are actively contributing to defining this protocol, representing a shared vision of multi-agent system
A2A operates on key principles like capability discovery (agents can publish their abilities via JSON-formatted “Agent Cards”), task management, collaboration, and user experience negotiation. Built on established web standards like HTTP and JSON, A2A ensures compatibility while prioritizing security. Google has released A2A as open source, encouraging community contributions.
Crucially, the sources position A2A as a higher-level abstraction for agent communication that complements Anthropic’s MCP. While MCP focuses on how agents interact with tools and data, A2A focuses on how agents interact with each other, regardless of their underlying technology.
Why These Protocols Matter: The Dawn of Interoperable AI
The emergence of MCP and A2A signals a critical turning point for the AI ecosystem. Their importance stems from several key factors:
- Breaking Down Silos: Both protocols directly address the fragmentation challenge. MCP provides a standardized way for any AI model to use any tool, while A2A aims to connect disparate AI agents.
- Accelerating Innovation: By reducing the friction of integration, these protocols free up developers to focus on building innovative applications and agentic workflows rather than wrestling with custom connections.
- Enabling Complex Automation: The ability for agents to seamlessly use a wide range of tools (via MCP) and collaborate with other agents (via A2A) unlocks the potential for highly sophisticated and autonomous workflows that were previously impractical.
- Fostering a Vibrant Ecosystem: Open standards encourage broader participation and the development of a diverse range of tools, clients, and agents. The emerging MCP marketplace is a testament to this.
- New Business Models: As every app potentially becomes an MCP client and every API a server, we may see new pricing models emerge based on dynamic tool selection and usage. The value proposition shifts from just having an API to having easily discoverable and effective tools for AI agents.
- Enterprise Adoption: For enterprises seeking to implement AI-driven automation across their diverse systems, both MCP and A2A offer pathways to achieve interoperability and scale their AI initiatives.
Investment Opportunities and the Road Ahead
From a venture capital perspective, the rise of MCP and A2A presents exciting opportunities:
- Companies building foundational MCP infrastructure and tooling: This includes server generation tools, hosting solutions, connection management platforms, and security solutions.
- Developers creating innovative MCP servers that unlock new capabilities for AI agents across various domains.
- Startups building next-generation MCP clients with compelling user experiences for both technical and non-technical users.
- Companies leveraging A2A to build sophisticated multi-agent systems for specific industry verticals.
- Platforms and marketplaces that facilitate the discovery, deployment, and management of both MCP servers and AI agents compatible with A2A.
The coming year will be pivotal. Will we see the rise of a unified MCP marketplace? Will authentication for AI agents become seamless? Can multi-step execution be formalised into the MCP protocol? Similarly, the adoption and evolution of the A2A protocol will be crucial to watch.
In conclusion, the Model Context Protocol and Google’s Agent2Agent protocol are not just incremental improvements; they represent a fundamental shift towards an interoperable AI future. By standardising how AI interacts with the world and with each other, these protocols are laying the groundwork for a new generation of autonomous, multi-modal, and deeply integrated AI experiences. For investors and builders alike, understanding and embracing this shift is essential to navigating the exciting landscape ahead.
I'm a former general partner at google ventures, head of Google for Entrepreneurs in Europe and founding head of Campus London, Google's first physical hub for startups.
I'm also the founder of Techbikers, a non-profit bringing together the startup ecosystem on cycling challenges in support of Room to Read. Since inception in 2012 we've built 11 schools and 50 libraries in the developing world.
- Weekly FIRGUN Newsletter – April 11 2025 - April 11, 2025
- The future of AI tooling is Interoperable: MCP and Agent2Agent - April 10, 2025
- Weekly FIRGUN Newsletter – April 4 2025 - April 4, 2025
