MCP server, official SDKs, pre-recorded audio, and Gemini 3.1 Live

Hey there 👋,

Four drops this month: an MCP server, official Python and TypeScript SDKs, hybrid pre-recorded audio in workflows, and Gemini 3.1 Live as a first-class speech-to-speech provider.

MCP server

Point Claude Code, Claude Desktop, or Cursor at your Dograh workspace. Your coding agent lists agents, reads their definitions, edits them in TypeScript, and saves drafts - without leaving the editor. The docs are indexed too, so "add a knowledge-base node that uses my PDFs" just works. One CLI command with your API key to register.

→ Connect your coding agent

Python & TypeScript SDKs

pip install dograh-sdk or npm install @dograh/sdk. Both are generated from the same OpenAPI spec, so parity is guaranteed. Build agents with typed wf.add() / wf.edge() calls, save as drafts (published agents keep serving), and trigger outbound calls programmatically. The MCP server is built on top of these.

→ Install the SDKs

Pre-recorded audio in workflows

Mix human recordings with TTS inside any node. Use a real clip for the greeting, disclaimer, or closing - keep the rest AI-generated. You cut TTS cost, shave first-token latency, and recover the emotional nuance synthetic voice still misses. Works with Cartesia, ElevenLabs, and Deepgram cloned voices; drop clips into prompts with @ syntax.

→ Add pre-recorded audio

Gemini 3.1 Live - speech-to-speech

Google's realtime multimodal model is now a first-class provider. One streaming connection replaces STT + LLM + TTS and shaves a meaningful chunk off turn latency. Default model gemini-3.1-flash-live-preview, five built-in voices (Puck, Charon, Kore, Fenrir, Aoede), English-only for now.

→ Configure Gemini 3.1 Live

How to access these

On app.dograh.com? You're already on the latest - nothing to do.

Self-hosted? Pull the new images and restart. The update guide covers both "latest" and pinned-tag flows for local and remote deployments.

→ Update your self-hosted stack

Community update

We just crossed 210 builders across 20 time zones in the community, and 375 stars on GitHub. Thanks for backing an open voice-AI stack - it makes a real difference at this stage.

If Dograh has saved you time, a star on GitHub helps the next person find us. We'd love to hit 1,000.

If you try any of these and something feels off - or if there's a model, provider, or integration you want next - hit reply. I read every message.

- Abhishek

Co-founder, Dograh

github.com/dograh-hq/dograh ⭐