Proposal: standalone vocagateway-mcp for private agent transcription

[neha-nupoor]

Context

Following discussion in Discord, I’d like to contribute a separate vocagateway-mcp repository: an MCP server for a user’s self-hosted VocaGateway.

Initial scope

• Gateway health/readiness and active-engine status
• Installed/available model listing
• Transcribe a local audio file through the user-configured gateway

The server would use the existing bearer authentication and make the gateway destination explicit before audio is sent. There would be no Voca-hosted service, account, telemetry, or secret/transcript logging.

API direction

Use the existing session API initially, with the OpenAI-compatible POST /v1/audio/transcriptions endpoint from #37 as the preferred simple transcription path once merged.

Non-goals for v1

• Streaming transcription
• Token administration
• Model download, deletion, or selection
• Any cloud relay or hosted gateway

Maintainer decision requested

If this scope looks right, I’ll create the standalone repository under the VocaHQ organization and open the initial implementation PR there.

[github-actions]

Thanks @neha-nupoor. A maintainer will look into it.

---

Meanwhile, connect with us:

[jatinkrmalik]

I am generally aligned with this, just a quick question, so this MCP is exclusively for the gateway right?

Because we have a vision of embedding the gateways also in the desktop apps so that you can flip it on with just one click and then use your primary machine as the gateway for all the other clients.

Which I think works really well here because we can also ship MCP along with that. It doesn't need to live in a same repository, but can be its own repository. And we can pull from the release tags.
In my opinion, containerization is also needed.

[jatinkrmalik]

@neha-nupoor I created https://github.com/VocaHQ/vocagateway-mcp for this. There's a short README on main so it isn't an empty repo. I sent you a maintainer invite on it. Open the first implementation PR there.