Costs and limits

Studio MCP uses the AI model supplied by your external client. Private MCP dispatch does not invoke VoxelSite’s built-in AI providers or its model Router, even when those providers are configured.

Keep the bills separate

A Codex or Claude subscription is not a promise of unlimited MCP work, and it does not fund VoxelSite’s separately configured provider API calls. Check your client’s current plan and usage limits directly.

External-only use does not require a built-in AI provider key. Choosing to use Studio generation later introduces that separate provider configuration and billing.

Work within returned limits

The site information tool reports authoring limits. In VoxelSite 1.33.0, core bounds include a 2 MiB MCP request body, up to 100 requested change operations and a 32 MiB managed draft byte budget for change validation. Individual text files in a change are limited to 1 MiB. These are ceilings, not recommended task sizes; hosting or client limits can be lower.

Reads, diffs, inventories and uploads also have bounded chunks or pages. Ask the client to follow the returned offsets, cursors and transfer limits. Do not paste an entire large site into one request or assume a truncated diff is complete.

A model-budget read reports local configuration and documented budget information. It does not contact a provider, measure your external client’s remaining subscription usage or guarantee that a model will accept the task.

Shared hosting still has limits

VoxelSite’s server runtime is PHP 8.2+ and SQLite. Buyers do not need Node.js or an npm build step. Your AI client runs separately and may have its own installation requirements.

Hosts can limit request size, execution time, memory and outbound connections. Check operation status after a timeout and continue the retained operation where instructed. Starting a duplicate operation is not a timeout remedy.

For a first edit, change one page or section, review the complete diff and preview it before increasing the scope.