# MultiVoice Studio plugin

The plugin lets ChatGPT and Codex parse speaker-labelled scripts, list your private consented voice profiles, save projects, check voice-credit usage, and create watermarked narration with an existing MultiVoice Studio account.

It does not sell subscriptions or credits. Account creation, plan selection, and billing remain on the MultiVoice Studio website.

## Tools

- `parse_script` — anonymous, read-only script parsing.
- `list_voices` — list the built-in model voice and the signed-in user's private voice profiles.
- `get_account_usage` — show current plan and remaining credits without promoting an upgrade.
- `list_projects` — list recent saved projects.
- `create_project` — save a script without generating audio.
- `create_narration` — generate one audio file and consume one credit per spoken character; returns an OAuth-protected MCP audio resource link.

## Authentication

The plugin uses OAuth 2.1 authorization code with PKCE (`S256`). Dynamic client registration and refresh-token rotation are supported. The protected resource is the deployed `/mcp` endpoint.

The repository package includes `.mcp.json`, so a Codex installation can discover the same production MCP endpoint directly; ChatGPT submission still uses the Platform portal's Universal MCP URL field.

## OpenAI submission checklist

The current OpenAI flow is a Platform plugin submission, not a legacy manifest upload. The owner must:

1. Verify an individual or business developer identity in the OpenAI Platform.
2. Have `Apps Management` write access (`api.apps.write`) for the publishing organization.
3. Create a plugin draft with the universal production MCP URL `https://multivoice-studio-production.up.railway.app/mcp` and scan its tools.
4. Supply the public website, support, privacy, and terms URLs, logo, category, starter prompts, and the test cases in [PLUGIN_TEST_CASES.md](PLUGIN_TEST_CASES.md).
5. Provide a reviewer account that works without MFA, email confirmation, SMS confirmation, or private-network access, then submit for review and publish after approval.

The MCP endpoint already exposes tool-level `securitySchemes`, OAuth discovery, PKCE, resource validation, and the required AI-generation disclosure. Billing remains on the website: the OpenAI product-checkout conversion is a beta program for approved partners, so this product does not expose a purchase tool in the initial submission.

## Example prompts

- Parse this dialogue and tell me which speakers you found.
- List my MultiVoice Studio voices, then cast this script.
- Generate this script as a multi-voice narration.