Integrate and evaluate: New MCP support and observability framework from AGNTCY

The AGNTCY has been working to build the foundational technologies and standards for multi-agent software development, allowing AI agents to discover each other and work together regardless of how they were built, who built them, or where they run.  

Today, I’m excited to announce several key updates from AGNTCY, including added interoperability for Anthropic’s Model Context Protocol (MCP) across several of our key components, a new observability data schema enriched with concepts specific to multi-agent systems as well as new extensions to the Open Agentic Schema Framework (OASF). 
 
All of these new developments are further building out the Internet of Agents—an open, interoperable internet for agent-to-agent collaboration.

Simplifying multi-agent software development with MCP interoperability 

MCP has gained significant attention and adoption recently, solidifying its position as an emerging open standard. As we discussed in this comparison analysis, AGNTCY’s Agent Connect Protocol (ACP) and MCP are complementary protocols: ACP enables autonomous agents to collaborate and share resources in a distributed system while MCP enriches individual AI models with external context to enhance their decision-making and response generation. We believe that both protocols are required for powerful and scalable multi-agent systems and have introduced new AGNTCY functionality to simplify developing multi-agent software using them together.

Scalable MCP communications using Agent Gateway  

Today, securely connecting large language models (LLMs) to a remote MCP server and managing distributed routing is fairly complex. It's currently handled through custom implementations which create tightly coupled applications that are especially fragile in a distributed environment. 

The Agent Gateway provides a low latency communication mechanism designed for the kind of multi-modal state exchange required for agentic communication, including for real time and interactive use cases. It supports both peer-to-peer and pub/sub (one to many) modes of communication.  
 
By adding support for MCP, AGNTCY’s Agent Gateway introduces new communication patterns between agents and AI models such as request-response, publish-subscribe, fire-and-forget, and streaming. Agent Gateway provides a scalable, streamlined solution that simplifies secure connections and distributed routing for MCP servers. 

How it works: Developers will be able to access MCP servers through Agent Gateway topics instead of connecting directly to MCP. Each MCP server will be exposed as a dedicated Agent Gateway topic, enabling message exchange over the data plane rather than JSON RPC. This feature allows multiple clients to share access without overloading the MCP server, improving scalability and aligning with our unified messaging infrastructure.

Extending API Bridge Agent to integrate with MCP 

The ability for agents to seamlessly interface with external services is critical for building intelligent, scalable multi-agent systems. The API Bridge Agent simplifies API access by acting as a dynamic translator between natural language commands and structured API calls, leveraging OpenAPI specifications. With the latest update, AGNTCY’s API Bridge Agent introduces support for MCP, unifying how agents interact with external resources, and further simplifying agent interactions with MCP servers. 

How it works: The API Bridge Agent selects the best resource to service an inbound query. After identifying the APIs or MCP servers that can handle the request, the API Bridge Agent converts natural language into the appropriate calls. Returned responses are transformed back into natural language.   

This unified approach provides greater flexibility and standardization for modern agent-based architectures allowing agents seamless access to external data and capabilities. To see the integration in action, please refer to the sample project here.

Introducing a new framework for Multi-Agent Software Observability & Evaluation 

In high-stakes scenarios, where even 95% accuracy can result in significant financial or security risks, AI application developers require deep end-to-end visibility into the performance of their multi-agentic systems. They must be able to evaluate the overall system's quality, validate its accuracy, and understand the decision-making processes. 

AGNTCY’s new observability and evaluation framework is designed to monitor and assess the performance of multi-agent software, providing visibility into the entire agentic workflow end-to-end. We will be publishing more on this vision soon.

Laying the Evaluate foundation with a new observability data schema 

To begin, AGNTCY has released a multi-agent system observability data schema based on Open Telemetry and following the LLM semantic convention. The schema is enriched with concepts specific to multi-agent systems such as agent collaboration success rate, system response time, and task delegation accuracy. We have established a working group with industry partners to help accelerate this standardization. You can follow it here. 

Advancing the Open Agentic Schema Framework: Agent Record Extensions 

The Open Agentic Schema Framework (OASF) has introduced new extensions to enhance its support for diverse agent frameworks, tools, technologies, and deployment patterns. These latest agent record extensions are specifically designed to integrate Evaluation & Observability data introduced in this release. 

With these updates, developers can now easily publish and share metrics, traces, events, and logs related to multi-agent software within the Agent Directory. This advancement streamlines collaboration and improves the accessibility of critical performance data for multi-agent systems. 

Help shape the future of the Internet of Agents 

If you are interested in adopting these protocols, tools, and frameworks, or even helping us shape the Internet of Agents, we’d love to hear from you. Contact us to get involved!