- AGNTCY is interoperable with MCP and A2A but brings more security features to bear on agentic AI
- It was originally developed by Cisco but is now overseen by the Linux Foundation
- Cisco exec Vijoy Pandey encouraged more companies to get involved in AGNTCY's development
If you’re working in the AI space, you’ve probably already heard of the model context protocol (MCP). Heck, you’re also likely familiar with Google’s Agent to Agent (A2A) protocol. But there’s one more tool that should be on your radar: AGNTCY.
Originally developed by Cisco, AGNTCY is a platform designed to provide the infrastructure AI agents need for discovery, communication, identity verification and observability.
In doing so, it fills in key security gaps in current inter-agent communications protocols like MCP. That, in turn, can help give enterprises the confidence they need to press ahead with more complex agentic AI deployments.
So, it’s kind of a big deal.
The project was recently open-sourced and now sits under the Linux Foundation’s collaborative umbrella. In addition to Cisco, Dell Technologies, Google Cloud, Oracle and Red Hat are working on AGNTCY as formative members of the project group.
How AGNTCY works
It’s worth noting that AGNTCY isn’t a replacement for MCP or A2A. Vijoy Pandey, SVP of Outshift at Cisco, told Fierce Network that AGNTCY is actually interoperable with both protocols as well as the Secure Low Latency Interactive Messaging (SLIM) protocol. That’s all on the comms and discovery side of the house.
Perhaps more importantly, though, AGNTCY brings security tools to bear in the agent-to-agent space. As Fierce noted previously, one of the key issues with MCP is that it lacks robust authentication and security features.
“Agents have human-like attributes but they operate at machine speed and scale. Everything that you’ve built so far has been built either for humans or machines,” Pandey said, explaining the problem agentic AI poses for security.
Among other things, AGNTCY provides cryptographically verifiable identities so that enterprises can ensure their agents are communicating with reputable outside agents. Then there are access controls to ensure the AI agents can only tap into the resources they need, when they need them. And if you read our recent piece on the role of access controls in mitigating enterprise security risks, you know this is important.
As Pandey explained, to keep AI agents in line, enterprises need to be able to grant and retract access authority at a task level. That’s already non-trivial to do but Pandey noted that even defining what constitutes a task can be hard.
“Those are two simple things that you take for granted that become complicated in an agentic environment,” Pandey said. “So, building out an identity framework and a service that, for example, deals with TBAC – task based access control – is something that we are doing through AGNTCY.”
Then there’s observability and evaluation, he said. AGNTCY is building tools that can help track whether agents are doing what they’re supposed to do and assign a reputation score for their performance.
What’s next
Going forward, AGNTCY’s direction will be determined by a steering committee comprised of individuals from the project’s formative member companies. But Pandey said he expects a strong focus on security, identity and trust.
“You’ll see a lot of focus going in that direction as the one burning thing that as a community we need to solve…for enterprises to embrace multi-agentic and agentic applications,” he said.
Indeed, Oracle’s SVP of AI and ML Roger Barga said in a press release accompanying the Linux Foundation announcement that “Enterprise customers need agent infrastructure they can trust for mission-critical workloads.”
Pandey concluded with a call to action, urging interested contributors, users and developers to step up to the plate to help push AGNTCY forward. You can learn more about the project at agntcy.org.