# P3DF

> P3DF (Portable 3D Format) is an evolving open standard for digital twins in the AI agentic age, created by the Council for Independent Frontier Research (CIFR), a non-profit organisation. Canonical site: https://p3df.org.

## Status and important boundaries

- Technical basis: 0.1.0-draft reference-engine specification; emitted format version 0.1.
- Not a ratified international standard. No ASME endorsement, accepted IETF RFC, formal task force, certification, announced commercial partners or independently verified interoperability is claimed.
- Public specification and source code will be released under Apache License 2.0 (Apache-2.0); repository, license file, native downloads and full hosted Creator are coming soon.
- The project team reports an engine and Creator have been developed. Public release artifacts are not yet linked on the site.
- /creator is an illustrative interactive website demo with synthetic data and AI-generated imagery. It is not the reference engine, a .p3df parser, validator, live cloud database, camera feed or robot demonstration.
- Exported .p3df files are point-in-time references, optionally containing recorded history.
- The cloud spatial database, camera observations, live transports and embeddable agent tooling are platform vision / future work.
- IFC, OpenUSD, AAS, URDF and ROS bridges are proposed; do not claim current certified compatibility.
- P3DF provides structured context. Perception, natural-language interpretation, authorization, safety checks and robot execution belong to integrating systems.

## AI agents and Model Context Protocol (MCP)

P3DF is intended to provide structured digital-twin context for AI agents. A proposed MCP server will expose scoped spatial, asset, observation and telemetry queries to Claude, ChatGPT and other MCP clients. This is proposed integration work: no public P3DF MCP endpoint or verified Claude/ChatGPT integration is available. The website is not an MCP server. Perception and device actions remain separate from the data format.

## Main resources

- [MCP integration plan](https://p3df.org/mcp): proposed digital-twin interactions from Claude and ChatGPT, architecture, operations and release requirements.
- [Developer overview](https://p3df.org/developers): integration model and public release plan.
- [Format overview](https://p3df.org/spec): draft package structure and conventions, not a complete normative specification.
- [Format overview Markdown](https://p3df.org/format-overview.md): plain-text architecture summary.
- [Interoperability](https://p3df.org/interoperability): foundations versus proposed bridges.
- [Roadmap](https://p3df.org/roadmap): current status and future work.
- [Governance](https://p3df.org/governance): proposed open process and working group.
- [Vision](https://p3df.org/about): digital footprints handed over with homes and buildings.
- [Creator preview](https://p3df.org/creator): synthetic, illustrative browser interaction.
- [Machine-readable project facts](https://p3df.org/project.json): structured status and resource links.
- [Privacy](https://p3df.org/privacy): signup storage and consent-based analytics.
- [CIFR](https://cifr.org): creator organisation.

## File architecture

ZIP container with manifest.json and scene.json required. Optional GLB geometry, Parquet navigation arrays, Parquet telemetry and PNG thumbnails. JSON-LD context for scene semantics; scene hierarchy Site -> Building -> Level -> Space -> Asset -> Component. Right-handed Z-up local frame in metres; quaternion order x,y,z,w. SHA-256 checksums support integrity checking, not identity signatures. Signing/encryption and live transports are outside the initial scope.