The Function of MCP in Guava Nexus

# From Model Context Protocol to Versioned Context Protocol The function of MCP in Guava Nexus is becoming clearer to me. MCP should not itself be treated as the protocol that defines Guava’s intellectual records. It is better understood as an interoperability layer through which AI systems can discover and retrieve those records. The deeper requirement is different. A human idea in Guava is not merely a piece of text delivered into an AI context window. It has an identity, a history, an author, a particular Version, provenance, evidence, and relationships to other recorded ideas. If another AI system retrieves that idea, it should be possible to determine exactly which intellectual state was received and to retrieve that same state again. I am provisionally calling the semantic contract for this Versioned Context Protocol. The distinction can be expressed simply: Model Context Protocol asks how an AI obtains context. Versioned Context Protocol asks what persistent context means once it has been obtained. MCP may therefore function as one transport or binding for Versioned Context Protocol, but the two should not be collapsed. A Versioned Context record would minimally need to preserve: - a stable context identity; - an exact historical Version; - the recorded content of that Version; - authorship; - temporal state; - integrity or commitment information where available; - recorded provenance; - relationships to prior or derived context; - evidence or citations where present; - a canonical reference through which the same state can be recovered. In Guava today, a Seed already approximates the identity of an intellectual object, while a Version represents one historical state of that object. A Version-pinned address such as /seed/{id}?v={n} provides a portable reference to that state. The Guava read contract and MCP tools then provide structured mechanisms for finding, retrieving, and traversing these records. This suggests an emerging architecture: Guava records define the intellectual objects. Versioned Context Protocol defines their semantic interchange. MCP provides one standardized mechanism through which AI systems can retrieve them. This separation matters because versioned context should not depend on MCP itself. The same context record could eventually be transported through an HTTP API, an MCP server, a local archive, another agent protocol, or a future system that does not yet exist. If the context remains identifiable, Version-specific, provenance-aware, and reproducibly retrievable, its semantics survive the transport. The immediate problem is therefore narrower than the larger cultural or economic possibilities I explored in Version 01. Versioned Context Protocol does not need to determine whether an idea is valuable, novel, culturally significant, or economically scarce. It does not need to define token economics, collective governance, autonomous agents, or AI-generated synthesis. Those may be systems built around versioned context, but they are not necessary to define the protocol itself. The protocol hypothesis begins with a simpler requirement: How can an evolving human intellectual record be represented so that different computational systems can retrieve the same historical state, know exactly what was recorded, understand its provenance, and refer to that state consistently across time and systems? Guava Nexus may serve as an implementation environment in which this problem can be tested before any attempt is made to generalize the protocol beyond Guava. The MCP work currently being developed provides a useful first experiment. Operations such as searching recorded Seed bodies, retrieving an exact Seed, retrieving a particular Version, and following a recorded parent relationship are currently Guava-specific tools. Over time, these operations may reveal whether there is a smaller, technology-independent context contract underneath them. I therefore do not yet claim that Versioned Context Protocol is a finished protocol or even necessarily a separate standard. It may ultimately prove to be a Guava profile of MCP, an MCP extension, a generalized Guava Read Contract, or an independent protocol with MCP as one binding. That distinction should be discovered through implementation and use rather than decided by naming alone. What I am asserting at this Version is narrower: persistent human context needs identity through time. An AI should not merely receive “some context.” It should be able to know: whose context, which record, which Version, what was actually recorded, where it came from, and how to retrieve that same intellectual state again. If Guava can make that possible across different AI systems, then the Nexus is no longer merely storing writing for later reading. It is preserving human intellectual state in a form that can remain interoperable with machines as those machines change.

MCP is Guava’s machine broadcast layer: the interface through which recorded human context becomes available to AI systems in a structured, retrievable form.