Is your feature request related to a problem?
Basis lowers the barrier to building social VR — worlds, games, interactive experiences. But "lower barrier" still assumes the creator can navigate Unity, understand a DevOps stuff, and write C#. For a lot of people with genuine creative vision, that's still a wall.
AI coding agents (Claude Code, Cursor, Antigravity, OpenCode, Kilo Code, Pi) are changing this. A creator who can describe what they want in natural language and iterate in conversation is increasingly able to build things they couldn't build alone. But these agents have no knowledge of Basis. Without framework-specific context, they fall back to generic Unity or XR Toolkit patterns that don't work here and produce broken, hard-to-debug results.
Concrete examples of what agents produce vs. what Basis actually requires:
"Make this object pickupable":
→ Agent writes XRGrabInteractable
→ Basis needs BasisPickupInteractable (with its own grab/throw physics, highlight feedback, and parenting logic) plus BasisObjectSyncNetworking and explicit ownership transfer — all tied to the framework's networked transform replication
There's also a tooling gap on the Unity side. Unity's built-in AI assistant requires a paid subscription and has inconsistent results with third-party frameworks. This leaves many creators — especially those without a budget — effectively without AI assistance inside their editor, even when they have a capable agent available in VS Code or a terminal.
Describe your preferred solution
A community-maintained basis-agentic-tools repository — a Basis equivalent of what Meta shipped for Quest developers with meta-quest/agentic-tools.
The core idea: a local MCP server paired with agent skills, installable in one command, working with any MCP-compatible agent — not tied to any single provider or paid subscription.
Repository structure:
basis-agentic-tools/
├── AGENTS.md # Agent navigation guide (canonical entry point)
├── CLAUDE.md # Symlink → AGENTS.md (Claude Code / Claude Desktop)
├── .mcp.json # Shared MCP server config for all supported clients
├── .claude-plugin/ # Claude Code plugin manifest
├── .cursor-plugin/ # Cursor plugin manifest
├── .github/plugin/ # GitHub Copilot CLI plugin manifest
├── mcp/
│ └── basis-mcp-server/ # Docs search, component registry, scene validation
├── skills/
│ ├── basis-new-world/ # Scaffold a Basis world from scratch
│ ├── basis-new-game/ # Scaffold a multiplayer game with Basis networking
│ ├── basis-interactable/ # Pickup, seat, pedestal — done right
│ ├── basis-networking/ # Object sync, ownership, state propagation
│ ├── basis-avatar/ # Avatar loading, IK calibration, pedestal setup
│ ├── basis-server-setup/ # VPS → Docker → .NET 9 server → monitoring
│ └── basis-code-review/ # Flag non-Basis patterns in existing Unity code
└── docs/
└── basis-components.md # Component reference served by the MCP server
How Meta's model works (and what we'd mirror):
- The MCP server provides tools: docs lookup, component registry, scene validation
- Skills are structured multi-step workflows that use those tools — e.g.
basis-interactable doesn't just list components, it walks the agent through the full pickup setup including ownership semantics
AGENTS.md is the single entry point for any agent regardless of provider; CLAUDE.md is just a symlink so Claude-family tools find it automatically
A creator without C# experience opens a Basis project, installs in one command, and prompts:
"Add a campfire players can sit around with a pickupable marshmallow stick"
The agent knows to use BasisSeat with correct IK points, BasisPickupInteractable + BasisObjectSyncNetworking for the stick, and handles ownership — instead of producing XR Toolkit boilerplate that compiles but does nothing.
Describe any considered alternatives
1. Just improve the written docs
Better docs help human readers, but agents don't browse docs.basisvr.org while working — they rely on context provided at session start. Docs without a delivery mechanism don't reach the agent.
2. A single AGENTS.md / CLAUDE.md in the main Basis repo
This is a minimal useful step and should happen regardless. But a flat file can't deliver multi-step workflow skills or expose live tools (scene inspection, component validation). It's a foundation, not a full solution.
3. Rely on Unity's built-in AI assistant
Unity now ships an AI coding assistant, but it requires a paid subscription and has inconsistent results with third-party frameworks like Basis. It's not a realistic option for indie creators or small teams without budget — precisely the audience most likely to benefit from AI assistance.
4. Rely on existing unity-mcp tools (CoplayDev or IvanMurzak)
Both are excellent general-purpose Unity MCP bridges (CoplayDev/unity-mcp, IvanMurzak/Unity-MCP) and could serve as the underlying layer. IvanMurzak in particular supports auto-generating skills from installed packages — worth exploring whether Basis SDK components could register themselves this way. But neither knows Basis-specific patterns out of the box; Basis skills and context would still need to be authored on top.
5. Wait for LLMs to learn Basis from public code
Basis is growing, but the training data window means agents will lag by months or years. Proactively providing context via MCP + skills is faster and more controllable than waiting for model updates.
Additional Context
This proposal is directly aligned with Basis's stated goal: equip VR creators and accelerate the growth of VR.
The democratization angle
The creator most likely to benefit from this isn't an experienced Unity developer — it's someone with strong creative vision (a game designer, an artist, a community organizer, a cultural institution) who has never written C# and is trying to build something in VR for the first time. These people are already using AI agents to write code. What's missing is the bridge between those agents and Basis.
The tooling stack for this already exists and is free: VS Code + Kilo Code (an open-source AI coding extension) connected to a capable free model is a genuinely functional starting point — no Unity subscription, no paid API key required. Add a Basis MCP server and skills, and that stack becomes able to scaffold a networked VR world or game with correct Basis patterns. That's a meaningful shift in who can build here.
This applies equally to worlds and games. Basis supports both — and the skill set needed to make a multiplayer VR puzzle game versus a social hangout space overlaps heavily at the framework level. Skills should reflect that.
The resilience angle
Creators who build cultural or community projects on closed Social VR platforms are exposed to platform shutdowns, regional access blocks, and policy changes outside their control. Basis, as MIT-licensed and self-hostable, is a meaningful alternative — but only if building on it is accessible enough for small teams and solo creators. Lowering the AI-assisted development floor directly expands who can build here and who can keep their work alive.
Structural reference:
Is your feature request related to a problem?
Basis lowers the barrier to building social VR — worlds, games, interactive experiences. But "lower barrier" still assumes the creator can navigate Unity, understand a DevOps stuff, and write C#. For a lot of people with genuine creative vision, that's still a wall.
AI coding agents (Claude Code, Cursor, Antigravity, OpenCode, Kilo Code, Pi) are changing this. A creator who can describe what they want in natural language and iterate in conversation is increasingly able to build things they couldn't build alone. But these agents have no knowledge of Basis. Without framework-specific context, they fall back to generic Unity or XR Toolkit patterns that don't work here and produce broken, hard-to-debug results.
Concrete examples of what agents produce vs. what Basis actually requires:
"Make this object pickupable":
→ Agent writes
XRGrabInteractable→ Basis needs
BasisPickupInteractable(with its own grab/throw physics, highlight feedback, and parenting logic) plusBasisObjectSyncNetworkingand explicit ownership transfer — all tied to the framework's networked transform replicationThere's also a tooling gap on the Unity side. Unity's built-in AI assistant requires a paid subscription and has inconsistent results with third-party frameworks. This leaves many creators — especially those without a budget — effectively without AI assistance inside their editor, even when they have a capable agent available in VS Code or a terminal.
Describe your preferred solution
A community-maintained
basis-agentic-toolsrepository — a Basis equivalent of what Meta shipped for Quest developers withmeta-quest/agentic-tools.The core idea: a local MCP server paired with agent skills, installable in one command, working with any MCP-compatible agent — not tied to any single provider or paid subscription.
Repository structure:
How Meta's model works (and what we'd mirror):
basis-interactabledoesn't just list components, it walks the agent through the full pickup setup including ownership semanticsAGENTS.mdis the single entry point for any agent regardless of provider;CLAUDE.mdis just a symlink so Claude-family tools find it automaticallyA creator without C# experience opens a Basis project, installs in one command, and prompts:
The agent knows to use
BasisSeatwith correct IK points,BasisPickupInteractable+BasisObjectSyncNetworkingfor the stick, and handles ownership — instead of producing XR Toolkit boilerplate that compiles but does nothing.Describe any considered alternatives
1. Just improve the written docs
Better docs help human readers, but agents don't browse docs.basisvr.org while working — they rely on context provided at session start. Docs without a delivery mechanism don't reach the agent.
2. A single AGENTS.md / CLAUDE.md in the main Basis repo
This is a minimal useful step and should happen regardless. But a flat file can't deliver multi-step workflow skills or expose live tools (scene inspection, component validation). It's a foundation, not a full solution.
3. Rely on Unity's built-in AI assistant
Unity now ships an AI coding assistant, but it requires a paid subscription and has inconsistent results with third-party frameworks like Basis. It's not a realistic option for indie creators or small teams without budget — precisely the audience most likely to benefit from AI assistance.
4. Rely on existing unity-mcp tools (CoplayDev or IvanMurzak)
Both are excellent general-purpose Unity MCP bridges (CoplayDev/unity-mcp, IvanMurzak/Unity-MCP) and could serve as the underlying layer. IvanMurzak in particular supports auto-generating skills from installed packages — worth exploring whether Basis SDK components could register themselves this way. But neither knows Basis-specific patterns out of the box; Basis skills and context would still need to be authored on top.
5. Wait for LLMs to learn Basis from public code
Basis is growing, but the training data window means agents will lag by months or years. Proactively providing context via MCP + skills is faster and more controllable than waiting for model updates.
Additional Context
This proposal is directly aligned with Basis's stated goal: equip VR creators and accelerate the growth of VR.
The democratization angle
The creator most likely to benefit from this isn't an experienced Unity developer — it's someone with strong creative vision (a game designer, an artist, a community organizer, a cultural institution) who has never written C# and is trying to build something in VR for the first time. These people are already using AI agents to write code. What's missing is the bridge between those agents and Basis.
The tooling stack for this already exists and is free: VS Code + Kilo Code (an open-source AI coding extension) connected to a capable free model is a genuinely functional starting point — no Unity subscription, no paid API key required. Add a Basis MCP server and skills, and that stack becomes able to scaffold a networked VR world or game with correct Basis patterns. That's a meaningful shift in who can build here.
This applies equally to worlds and games. Basis supports both — and the skill set needed to make a multiplayer VR puzzle game versus a social hangout space overlaps heavily at the framework level. Skills should reflect that.
The resilience angle
Creators who build cultural or community projects on closed Social VR platforms are exposed to platform shutdowns, regional access blocks, and policy changes outside their control. Basis, as MIT-licensed and self-hostable, is a meaningful alternative — but only if building on it is accessible enough for small teams and solo creators. Lowering the AI-assisted development floor directly expands who can build here and who can keep their work alive.
Structural reference: