x-grok

This server exposes the Voice assistant's provider connection to your agents as 6 grok_* tools: text chat, image generation, text-to-speech, short-video generation, live web and X search, and an authentication status check. It is the same Grok (xAI) connection Voice runs on — one provider setup, two consumers: the hands-free assistant you talk to, and any agent session that wants a Grok capability mid-task.

The provider model is worth stating plainly, the same way the Voice page does: this does not run on your agent subscription. You bring your own Grok subscription (an in-app sign-in) or an xAI API key, and usage rides on that account under its own limits — the platform adds no charge on top. Because search grounding and media generation come with many Grok plans, agents often treat this server as the low-friction path for "what is being said right now" questions and quick illustrative media, keeping the agent session's own context for the actual work.

One naming note so the install makes sense when you look at it: the server is registered under the key x-grok, but its package sits at mcp-servers/voice/ — named for the feature it powers rather than the provider it talks to. It is one of only two servers whose key and folder differ (email, in email-server/, is the other), so a search by key can come up empty on a directory that is plainly there.

How your agents use it

• "What is X saying about the model release today?" — grok_x_search returns live X and web results the agent can cite, without any harvesting setup.
• "Give me a quick illustrative image for this draft." — grok_image; for narration or a short clip, grok_tts and grok_video.
• "Is the provider connected?" — grok_auth_status before a media-heavy task, so a missing sign-in surfaces early.

Prerequisites

A connected Grok account or xAI API key, set up in the Voice view's Connections menu — the Voice page covers the provider reality in full, including what works without any provider at all. Key storage follows the platform's bring-your-own-keys model (see the configuration reference).

Subscription — Ultra, because Voice is an Ultra capability. The gate is partial and the split is worth knowing: the five tools that actually spend your provider account — chat, search, image, speech, video — refuse on a Pro install, while grok_auth_status keeps answering, so a session can always find out where it stands before it tries anything. Your own provider subscription is a separate matter again: it pays for the calls, and Ultra is what makes the tools reachable. Tiers covers what each tier includes.

Tool reference

Tool	Parameters	What it does
grok_auth_status	—	Introspect the local x-grok OAuth state. Returns whether tokens are stored, when they expire, and the account email (if available). Use to debug auth issues or check before a long-running flow.
grok_chat	prompt*: string, system: string, model: string	Send a single-turn prompt to Grok 4.3 (xAI) via the SuperGrok subscription. Returns the model's text response. Use when you want Grok's perspective specifically — e.g. for X/Twitter-aware queries, a non-Claude/non-Gemini second opinion, or content that benefits from Grok's training distribution. 1M-token context window.
grok_x_search	query*: string, allowed_x_handles: array, excluded_x_handles: array, from_date: string, to_date: string, enable_image_understanding: boolean, enable_video_understanding: boolean, model: string	[Phase 2 — not yet implemented] Ad-hoc X (Twitter) search via xAI's built-in x_search Responses API tool. Returns Grok's grounded answer plus citation URLs. Use for current X-side discussion, reactions, or claims — NOT for bulk harvesting (observer/Nitter is the harvest path).
grok_image	prompt*: string, aspect_ratio: string, resolution: string, model: string	[Phase 2 — not yet implemented] Generate an image with Grok Imagine. Saves the result to data/images/ and returns the absolute path.
grok_tts	text*: string, voice_id: string, language: string, codec: string	[Phase 2 — not yet implemented] Synthesize speech with xAI TTS. Saves audio to data/audio/ and returns the absolute path.
grok_video	prompt*: string, image_url: string, reference_image_urls: array, duration: integer, aspect_ratio: string, resolution: string, model: string	[Phase 3 — not yet implemented] Generate a video with Grok Imagine Video. Text-to-video or image-to-video (up to 7 reference images). Includes diegetic audio. Saves to data/videos/ and returns absolute path.

Where to go next

• Voice — the feature this provider surface powers
• BYO Claude & keys — how Voice's provider model differs from the agent fleet's