We're Joining the Agentic AI Foundation (AAIF) to Help Open Protocols Work for Science

A key component of the future of AI-accelerated science is the way in which the intelligence layers talk to each other, to data, and to the universe of tools. Today, the Model Context Protocol (MCP) is the de facto standard that articulates that interaction. As MCP is the central artery of our main product, Valency Bond™, we want to support its maturation in the service of science. To that end, I am very excited to announce that Valency has joined the Linux Foundation’s Agentic AI Foundation (AAIF), focused on the MCP project.

Anthropic donated the protocol to the AAIF in December 2025, placing it under the same neutral Linux Foundation (LF) stewardship that carries Kubernetes, Node.js, and PyTorch, with Google, Microsoft, AWS, Cloudflare, and Bloomberg behind it. I believe that the MCP project hosted within the Linux Foundation is a great collaboration anchor. Back in 2020, my IBM edge computing team was a founding member of LF Edge, and I had the opportunity to serve on the LF Edge Board and to collaborate with the great folks at the Linux Foundation. In my role on the LF Edge Board, I saw how big companies could be great stewards, but also how critical it is for projects to be guided by all stakeholders. Joining AAIF now as a small company, we are excited to have a seat at the table as Valency works to shape the agentic AI and MCP future.

Why does Valency care about open protocols and MCP?

Valency is building the data layer for researchers and scientists working with AI. We need to move search results built from trusted scientific literature and research context to whatever model a researcher uses for the intelligence layer. MCP is the critical connective tissue for all the surfaces and operation modes we support: Claude, ChatGPT, Google’s Antigravity, or self-hosted open-weight models running on-prem, on-laptop, inside a university cluster, or behind a national lab firewall.

A lot of our work internally goes into the normalization, cleaning, and deduplication across all the research data sources we serve. MCP allows our backend to converse with the vast heterogeneity of the LLM surfaces, building software against an abstracted universal protocol. Thanks to MCP, we’re able to optimize for the things intelligence layers appreciate from their tool calls: response speed, parallelism, secure connections, verbose and informative error messages, suggestions for next best action, token efficiency, and — of course — correctness.

What does success look like for open protocols?

MCP is a new project and a new idea.¹ And it’s evolving quickly, with the latest incarnation of the specification that came out last month. It is changing in response to a dynamic and growing ecosystem. We’re excited for a seat at the table to help ensure agentic AI works for science going forward.

Recently my co-founder Josh wrote about “the invisible future of AI,” noting that the more successful and ubiquitous an underlying technology becomes, the less visible it is. In time, if the community is successful, MCP will be less of a headline grabber, and can settle in to quietly deliver “seamless integrations” to the global community — serving the greater good in the way that IEEE 802.11, Linux, and TCP/IP have done.

Journeys like this one are better together, with a diverse cohort. That was true for Linux when it started thirty-five years ago, and it’s true for agentic AI and MCP now.

Come Find Us in Amsterdam!

Valency will be at the largest MCP dev conference in the world, AGNTCon + MCPCon Europe in Amsterdam, September 17–18. We’ll have a booth and we’ll be demoing Valency Bond MCP. Come on by and tell us what you’re working on — and what you’d like to see more of! We’d also like to hear about your use cases around science and research.