What is VCP?

A protocol for portable, adaptive, live context across AI and services.

Every new AI service starts from zero, and to be understood you have to overshare. VCP is an open protocol that sends the what and keeps the why. You declare your preferences, constraints, and current state once. Services and agents that speak VCP receive short flags, plain enough for you to read, that they can honor without learning your story.

Three properties that matter

Today your context is stuck inside each service, the same in every situation, and frozen at whatever you last entered. VCP makes it portable, adaptive, and live.

Portability

Define your context once. Compatible services can read the fields they support, so you repeat yourself less.

Adaptation

Context can switch with your situation, from work mode at the office to personal mode at home, when the integration activates the matching profile.

Liveness

Tell a service you are tired or pressed for time, and where it supports live updates, the tone and depth of its answers can shift with you.

How it works

You declare your context. VCP encodes it as structured data. Services read supported flags and adapt without seeing private reasons behind them.

You set context

Define your goals, constraints, and current state. Mark what is shareable and what stays private.

VCP encodes it

Private circumstances become categorical flags. "Thin walls" becomes noise_restricted. The flag travels. Your story stays home.

Services adapt

Apps and services that speak VCP read the flags they support and adjust. They can update as your state changes.

The protocol ships as a specification, not a promise. VCP defines how context is written, carried, and checked; it cannot make a service obey it. Only a service that implements VCP reads the flags, and that service decides what to do with them. A signature shows which trusted key signed a bundle and that it arrived unchanged, not that anyone followed it. And a flag says far less than your story, but not nothing: consent, storage, and what a service infers from it remain the job of the application that holds your data.

Topology · I-T-S-A-M-E

Six protocol layers.

L1 Identity naming, namespaces, registry, encoding WHO & WHAT?

L2 Transport signed bundles, content hashes, signature verification HOW?

L3 Semantics CSM-1 grammar, personas, traits WHAT?

L4 Adaptation context encoding, state, decay WHEN & HOW?

L5 Messaging inter-agent exchange, escalation WHO TALKS?

L6 Economic Governance fiduciary constraints, authorization gaps, transaction governance WHO PAYS?

IDENTITY TRANSPORT SEMANTICS ADAPTATION MESSAGING ECONOMIC GOVERNANCE

Protocol layer topology

Three context tiers

VCP carries context in three tiers that change at different speeds: stable rules (the constitutional tier), shifting situations (the situational tier), and moment-to-moment state (the personal tier).

Constitutional rules

What the AI should and should not do.

Personas, safety boundaries, rulesets, adherence levels

Changes rarely

Situational context

Where, when, who, what occasion.

Morning vs. evening, alone vs. with children, home vs. work

Changes per session

Personal state

How you are right now.

Cognitive state, emotional tone, energy level, perceived urgency, body signals

Changes moment to moment

Not a magic mirror

You declare your context, and the AI adapts to what you share. If a system infers anything, VCP requires it to say so, show its confidence, and let you correct it.

Common approaches

VCP approach

AI silently infers your state

You declare your state; anything inferred is labeled

Context locked to one platform

Context travels to platforms that speak VCP

Static profile, set and forget

Live state, updated where integrations support it

Platform decides what to share

You control what is shared

Works with MCP

VCP is designed to work with the Model Context Protocol (MCP). MCP is the delivery channel. VCP is the structured content it carries.

MCP

Delivery

Connects AI agents to tools and data sources. The plumbing.

VCP

Policy

Encodes values, context, and constitutional rules. The meaning.

The Python SDK includes an MCP server, so an MCP-compatible agent can call VCP tools to validate a token, parse a CSM-1 code, and encode the current context. In browsers that support WebMCP, the SDK's WebMCP package registers VCP tools on the page itself.

See it in action

These are six of the site's nine demos, each a scripted simulation. Three show portability, adaptation, and liveness; three take VCP into energy negotiation, school governance, and medical diagnosis.