The open standard for encoding values in AI context.
MCP moves data. VCP encodes what matters about that data.
Overview | Architecture | Quick Start | Extensions | MCP Bridge | Governance | SDKs
The Value-Context Protocol (VCP) is an open specification for transporting constitutional values, behavioral rules, and personal context to AI systems.
AI systems accept text input but have no native ability to resolve references, verify signatures, or validate behavioral constraints. VCP provides a signed envelope format with cryptographic verification at the orchestration layer, delivering complete, self-contained behavioral context to the model.
| Property | Description |
|---|---|
| Portability | Define your context once — every compatible service receives it |
| Adaptation | Context profiles shift by situation: work mode, personal mode, crisis mode |
| Liveness | Real-time personal state (energy, focus, urgency) modulates AI behavior |
| Verification | Cryptographic signatures and content hashes ensure integrity |
| Privacy | Share influence without sharing information |
| Extensibility | Stable core + opt-in extensions for specialized needs |
VCP and MCP (Model Context Protocol) are complementary:
- MCP provides standardized transport for AI tool integration — tools, resources, prompts, sampling
- VCP provides standardized value transport — constitutional profiles, personal state, behavioral constraints
The VCP-over-MCP bridge makes this concrete: VCP tokens travel as MCP resources, VCP operations are exposed as MCP tools, and VCP context is injected into MCP sampling requests.
VCP is a six-layer protocol stack — I-T-S-A-M-E ("It's-a me!"):
┌──────────────────────────────────────────────────────┐
│ Layer 6 — VCP/E ECONOMIC GOV WHO PAYS │
│ Layer 5 — VCP/M MESSAGING WHO TALKS │
│ Layer 4 — VCP/A ADAPTATION WHEN and HOW │
│ Layer 3 — VCP/S SEMANTICS WHAT values mean │
│ Layer 2 — VCP/T TRANSPORT HOW values travel │
│ Layer 1 — VCP/I IDENTITY WHO and WHAT │
├──────────────────────────────────────────────────────┤
│ Core Security │
│ Encryption | Scanning | Opacity | Revocation | Audit │
├──────────────────────────────────────────────────────┤
│ Extensions (VCP-X-*) │
│ Personal | Relational | Consensus | Torch | Intent │
└──────────────────────────────────────────────────────┘
The six layers:
- VCP/E — Economic Governance: Fiduciary constraints, authorization gaps, transaction governance
- VCP/M — Messaging: Inter-agent message types, escalation severity, delivery semantics
- VCP/A — Adaptation: Context dimensions, state machine, hooks
- VCP/S — Semantics: CSM-1 constitutional encoding, personas, composition
- VCP/T — Transport: Signed bundle format, manifests, trust anchors
- VCP/I — Identity: Token format, namespace tiers, identity encoding
Core security (v3.1): Context encryption, injection scanning, context opacity, revocation infrastructure, tamper-evident audit chain.
Extensions (v3.1): Opt-in protocol extensions negotiated per session via capability handshake.
VCP Newcomer Guide — What VCP is and why it exists.
See how VCP works end-to-end with annotated examples:
- CSM-1 Encode/Decode — Token lifecycle
- Personal State Roundtrip — Signal encoding + decay
- Capability Handshake — Extension negotiation
Python:
pip install creed-sdkTypeScript:
npm install @creed-space/vcp-sdkRust:
[dependencies]
vcp-sdk = "3.1"from creed_sdk import VCPClient
client = VCPClient()
token = client.create_token(
persona="supportive_companion",
dimensions={"empathy": "high", "transparency": "high"},
scope={"domain": "mental_health"}
)
print(token.csm1) # csm1:supportive_companion:EH-TH-...VCP v3.1 introduces five protocol extensions:
| Extension | Status | Description |
|---|---|---|
| VCP-X-Personal | Stable | 5 personal state dimensions with intensity + configurable decay |
| VCP-X-Relational | Stable | Trust, standing, norms, AI self-model with mandatory uncertainty |
| VCP-X-Consensus | Stable | Schulze voting for multi-stakeholder constitutional deliberation |
| VCP-X-Torch | Stable | Session handoff for relational continuity across agents |
| VCP-X-Intent | Experimental | Transparent, correctable intent inference from personal state |
Extensions are opt-in and negotiated per session via capability handshake. Each extension has a spec, JSON schema, and wire format examples.
See the Extension Model overview for architecture details.
VCP layers map to MCP primitives:
| VCP Layer | MCP Primitive |
|---|---|
| Identity | Tools (vcp_validate_token) |
| Transport | Resources (vcp://bundle/*) |
| Semantics | Tools + Resources |
| Adaptation | Resources + Sampling integration |
| Messaging | Tools (vcp_send_message, vcp_escalate) |
| Economic Governance | Tools (vcp_authorize_transaction) + Resources |
| Extensions | Negotiation-filtered tools + resources |
The bridge enables any MCP-compatible client to access VCP values without implementing VCP natively. See the MCP Bridge spec.
VCP is a six-layer protocol stack (I-T-S-A-M-E):
Layer 6 -- VCP/E ECONOMIC GOV WHO PAYS
Layer 5 -- VCP/M MESSAGING WHO TALKS
Layer 4 -- VCP/A ADAPTATION WHEN and HOW constitutions apply
Layer 3 -- VCP/S SEMANTICS WHAT the values mean
Layer 2 -- VCP/T TRANSPORT HOW values travel securely
Layer 1 -- VCP/I IDENTITY WHO and WHAT is being addressed
VCP v3.1 Specification — Master specification referencing all layers, extensions, and security features.
| Document | Description |
|---|---|
| VCP Specification v1.0 | Full protocol specification |
| VCP v1.1 Amendments | R-line, personal state additions |
| Academic Paper | Formal paper |
| VCP/M Messaging v2.0 | Inter-agent messaging and escalation |
| VCP/E Economic Governance v2.0 | Economic governance layer |
| VCP/C Competence v2.0 | Competence assessment and adaptive friction (Supplementary) |
| Layer | Documentation |
|---|---|
| VCP/I — Identity | Naming, Namespace, Encoding |
| VCP/T — Transport | v1.0 Spec SS6 |
| VCP/S — Semantics | CSM-1 Grammar, Composition |
| VCP/A — Adaptation | Adaptation, Context |
| VCP/M — Messaging | Specification |
| VCP/E — Economic Governance | Specification |
Encryption, Injection Scanning, Context Opacity, Revocation | Audit Chain
| Layer | Status | Documents |
|---|---|---|
| VCP/I — Identity | Stable | 5 docs |
| VCP/T — Transport | Stable | 1 spec + 1 amendment |
| VCP/S — Semantics | Stable | 4 docs |
| VCP/A — Adaptation | Stable | 4 docs |
| VCP/M — Messaging | Stable | 1 spec |
| VCP/E — Economic Governance | Stable | 1 spec |
| Document | Description |
|---|---|
| Naming | UVC naming conventions |
| Encoding | Encoding formats |
| Ontology | Value classification |
| Registry | Registry protocol |
| Governance | Namespace governance |
See CHANGELOG.
| Schema | Validates | Version |
|---|---|---|
| vcp-manifest-v1 | Bundle manifests | v1.0 |
| vcp-identity-token | Identity tokens | v1.0 |
| vcp-semantics-csm1 | CSM-1 tokens | v1.0 |
| vcp-adaptation-context | Adaptation context | v1.0 |
| vcp-capability-handshake | Capability negotiation | v3.1 |
Extension schemas are co-located: specs/extensions/VCP-X-*/schema.json.
VCP is governed by a Technical Steering Committee (TSC) under a foundation-compatible model:
- Decision process: Lazy consensus, supermajority for spec changes
- Contribution: DCO (Developer Certificate of Origin) sign-off
- Extension proposals: Via the VEP process
- IP license: MIT License with Agentic AI Foundation transfer clause
See GOVERNANCE.md for the full charter and CONTRIBUTING.md for contribution guidelines.
| VEP | Title | Status |
|---|---|---|
| VEP-0001 | Extension Model Architecture | Accepted |
| VEP-0002 | Capability Negotiation Protocol | Accepted |
| VEP-0003 | VCP-over-MCP Bridge | Accepted |
Reference implementations live in the VCP-SDK repository:
| Language | Version | Status |
|---|---|---|
| Python | 3.1.0 | Reference implementation |
| Rust | 3.1.0 | High-performance / WASM |
| TypeScript | 3.1.0 | Browser-side |
- VCP Inspector: Interactive web tool for decoding/encoding VCP tokens — Live at inspector.valuecontextprotocol.org
- VCP Examples: 10 runnable examples covering all extensions (planned)
- Anti-drift CI: Automated spec-implementation version sync checks
- Agentic AI Foundation: Transfer of VCP governance to neutral foundation
We welcome contributions. See CONTRIBUTING.md for guidelines.
- Typo / Clarification: Open a PR directly
- Spec Change: Follow the VEP process
- Extension Proposal: Use the VEP template
Please read our Code of Conduct before participating.
This project is licensed under the MIT License.
Nell Watson, Elena Ajayi, Filip Alimpić, Awwab Mahdi, Blake Wells, Claude (Anthropic)
A Creed Space project, developed for contribution to the Agentic AI Foundation.