Scaling multi-agent interactions requires a robust, standardized interoperability layer – especially as the enterprise AI landscape continues evolving. To orchestrate everything from supply chain logistics to customer support, agents need a reliable way to discover one another, maintain context across distributed systems, and communicate asynchronously to execute complex workflows autonomously.
We are excited to announce the general availability of the MuleSoft Agent2Agent (A2A) Connector on MuleSoft’s Anypoint Platform. These powerful new capabilities build on Google’s open-source A2A protocol, enabling AI agents to securely communicate, collaborate, and take coordinated action across complex enterprise environments. This release brings enhancements that introduce critical standards for agent identity, governance, and event-driven orchestration.
Key capabilities of the A2A Connector
Building on the beta version of the connector, we have made architectural changes in this version. Here are the key capabilities introduced:
Bidirectional interoperability
The Connector abstracts the complexity of the protocol, handling the handshake and formatting required by the A2A specification. The connector operates not just as a Client consuming agent services, but as a fully compliant Server, effectively democratizing the agent economy. By deploying the A2A Connector as a server, existing assets (e.g. legacy APIs, integration flows, or custom microservices) can be wrapped and exposed as A2A-compliant agents.
Streamlined agent discovery and standardization
For this ecosystem to function at scale, discovery must be standardized. Whether an agent is a native AI service or a wrapped Mule flow, it must identify its capabilities and protocol compatibility before initiating a task. To facilitate this, we have aligned with the updated, formalized A2A specification for Agent Cards, allowing for discovery and retrieval of the well- known agent card. Additionally, the protocol version parameter allows agents to explicitly declare the version of the A2A specification they support (e.g. 0.3.0), ensuring compatibility and preventing runtime errors during handshakes between disparate systems.
Sync, stream, and resubscribe
Real-world workflows require diverse communication styles. We have expanded the connector to support the full spectrum of interaction:
- Synchronous messaging: The new Send Message operation provides a direct request-response capability for immediate, transactional queries.
- Resilient streaming: The connector now supports multimodal streaming, allowing for lower-latency data transfer between agents and enabling real-time token generation and a more responsive overall system.
- Connection recovery: This allows the client to seamlessly reconnect without restarting the task should a connection drop during a stream.
Asynchronous, event-driven architectures
An effective multi-agent system requires agents to react to state changes – such as a task completion or a data update – without inefficient polling. To support this asynchronous pattern, the connector’s Push Notification capabilities have been enhanced to include:
- Comprehensive configuration management: This enables agents to dynamically manage their subscriptions to external events.
- Server-side validation: This allows the receiving agent (even a “wrapped” flow) to validate incoming configuration requests before routing them, ensuring that only authorized and valid subscriptions are accepted.
- Protocol compliance: To ensure stability, we have introduced specific error handling for environments where these capabilities are not yet active.
Strengthening governance and contextual integrity
As agents begin to autonomously trigger actions within the enterprise, maintaining security boundaries and context is paramount. Compliance relies on answering critical questions like, “Who initiated this request?” and “Which workflow does this belong to?”. To address this, we have enhanced the metadata and attribution layers:
- Client identification: This enables the Agent Fabric to correctly identify the requestor and apply appropriate governance policies.
- Context propagation: This allows organizations to audit the full lifecycle of a request as it traverses multiple agents.
- Task cancelation: This allows an agent or admin to halt an ongoing process immediately, ensuring control over long-running workflows.
A foundation for the agentic enterprise
The transition to an agentic enterprise is as much an integration challenge as it is an AI challenge. By acting as a bidirectional bridge, the MuleSoft A2A Connector allows organizations to leverage their existing digital assets while adopting new standards for discovery, event listening, and governance. Watch our demo to learn more about the MuleSoft A2A Connector.
