As protocol standards mature, are enterprises ready for agentic AI?
There has been significant progress in AI in terms of generative AI (GenAI), agentic AI and growing interest in physical AI. While hardware, software, frameworks and AI ecosystems are rapidly evolving, innovation is clearly outpacing adoption. This is reminiscent of the late 1990s when the internet entered mainstream consciousness. Enterprise adoption was initially limited by perceived benefits of lower costs, increased efficiency and access to new markets.
Similar to the internet’s transformative effects, agentic AI will change how businesses derive intelligence and pass on the benefits to their customers. Organizations clearly understand that AI improves efficiency but remain concerned about adoption costs. Developers are overwhelmed by the wide variety of available frameworks, tools, models and concepts, and at the same time, struggle with fundamental ways to orchestrate their applications with enterprise data and existing intelligence. The core challenge remains: how to effectively connect tools, data sources and agents together to deliver intelligence to customers.
Recently, there has been significant buzz around protocols to help developers adopt agentic AI and multi-agent systems. The introduction and evolution of Protocolo de Contexto Modelo (MCP), Agent Communication Protocol (ACP) and Agent to Agent Protocol (A2A) signals a new era where agents can effectively collaborate, exchange information, access tools and deliver intelligence across cloud and edge environments.
MCP: Hype or Hope?
MCP has generated significant discussion. Over the past few years, we’ve seen major innovation in AI platforms and infrastructure. While RAG and function calling have improved AI interactions, building AI apps or agents remains challenging for developers. This is why MCP is emerging as a pivotal standard, offering AI developers a seamless way to interact with downstream services and simplify context building. Context building with relevant data is the core to creating high-quality agents. MCP addresses these needs and enables LLM and services interaction. It is like the Thunderbolt, HDMI and DisplayPort type of protocol that enables efficient communication for different purposes.
MCP substantially simplifies agentic AI adoption for developers. This roadmap created by the MCP community clearly defines priorities and direction, providing helpful guidance for implementation. Organizations will also benefit from the key initiatives outlined in the roadmap, like the MCP Registry, which enables developers to build a comprehensive network of agents. The emergence of OAuth as a complementary standard protocol strengthens agent ecosystems even more.
As with any other framework, MCP has its challenges. MCP offers a wide array of tools to support LLM reasoning, but it doesn’t prioritize coordinated, high-quality task execution. Developers may have limited control over tool usage, relying heavily on the LLM’s discretion. It’s important to recognize that MCP isn’t a one-size-fits-all solution — careful tool integration and thoughtful prompt engineering are essential for achieving high-quality outputs. Another concern is security. The persistent context, long-lived sessions and structured prompts could potentially introduce security challenges that must be addressed early in system design. Scalability is also a concern but as technology evolves, vendors will add incremental support to make it easier to scale.
ACP: The Local Collaboration Enabler
ACP is an open standard designed to enable seamless communication between AI agents, regardless of their internal technology and implementation. It provides standardized RESTful APIs for managing and executing agents, supporting both synchronous and asynchronous interactions. ACP focuses on interoperability, allowing agents from different technology stacks to collaborate effectively. It stands out among other communication protocols by enabling seamless interaction between autonomous agents, optimized for local-first setups like clusters or laptops running multiple cooperating agents. This protocol resembles Android’s Intents for interacting with other apps, or iOS’s Universal Links and Custom URL Schemes to help developers facilitate complex system interactions locally.
A2A: Breaking Down Cross-Platform Barriers
Google’s A2A protocol is an open standard designed to enable seamless communication and collaboration between autonomous AI agents across different frameworks or vendors. It focuses on interoperability for information exchange, coordination and collaboration across diverse enterprise platforms and applications. With the comprehensive approach to agent discovery, task management and secure collaboration, A2A represents a significant leap forward.
A2A is a boon for developers to build modular AI systems that work across platforms and enterprises. This is a major shift from the current approach, reducing vendor lock-in and enabling richer, cross-domain solutions. For example, an enterprise AI agent handling logging quality could coordinate with separate agents to build operational analytics using different software stacks. In essence, A2A has the potential to be a fundamental protocol similar to HTTP to power agents across the globe.
What Does All This Mean for Developers?
This is all exciting news for developers. Until now, they’ve shouldered the burden of agent construction from scratch. These new protocols ease this burden. However, the fruits of all the AI innovation only become a reality when complex use cases that involve integrating data and systems across multiple domains become easier.
MCP will be the driving force behind exposing functionality going forward. It will enable interactive access to domain and business data in a structured manner for developers to build high quality agents. MCP will also enable pulling in data from diverse sources such as sales data, knowledge bases, Wikipedia, scientific data and more to help agents solve real-world problems. In addition, MCP will simplify prompt engineering, allowing servers to provide templates more suitable for their specific domain and helping developers to construct prompts more easily than before. Perhaps most significantly, LLMs will no longer be constrained by stale training data, instead accessing fresh, diverse information through MCP servers.
ACP will make it easier to implement AI agents on edge and local devices. In instances where the majority of decision-making happens “on the go” in a disconnected environment, this protocol will be useful. Now, developers can build modular systems that can coordinate with a standard protocol to make edge AI easier.
A2A will gain momentum and enable cross-platform agents to work together to deliver superior intelligence to customers. A2A will help coordinate agents built using diverse frameworks with a common standard. The main requirement for this is to build an Agent Card that allows agents to be used and consumed by others.
These three protocols will complement each other: use MCP to build agents, ACP to extend them locally and A2A to extend them across network boundaries as described in the diagram below.
What Does All This Mean for Software Vendors?
The AI space is evolving significantly and software vendors are struggling to keep up with innovation. A lack of standard protocols made ROI questionable and building precise business cases for customers difficult. Large vendors like Amazon and Microsoft have had the resources to match innovation speed, while medium software vendors have played a “wait and see” approach. New startups typically focused on niche use cases to push their identity, though there were no guarantees as to whether their solution would last. These emerging protocols finally provide the standardization needed to reduce risk and accelerate adoption.
There has been significant adoption of MCP resulting in several vendors delivering MCP servers. Couchbase lanzado a version of MCP designed to support AI agentic workflows and applications by enabling LLMs to take actions on Couchbase clusters using a structured set of tools. Expect more innovation on this front in the future. The industry has also seen implementations of MCP mesh to enable a network of decentralized LLMs, agents and processing nodes to exchange information.
A2A-type standards emerged from vendors like Amazon, OpenAI and Microsoft, including Agent Network Protocol (ANP) and Agent Discovery Protocol (ADP). Organizations naturally want to influence protocols to use their own services. However, this could potentially fragment the protocols and slow adoption. This fragmentation will likely lead to a new middleware ecosystem, with startups stepping in to bridge the divides. Fast-moving software vendors will have greater success making their services available to customers.
As the developer data platform that empowers organizations to effectively use AI, Couchbase is excited about the ongoing development of these protocols and standards. We are prepared to ride this wave of innovation by making our data stores integrate seamlessly with these protocols to help customers build intelligent AI systems.