grok-oauth-mcp

grok-oauth-mcp

Enables local MCP clients to interact with xAI Grok through OAuth PKCE, supporting chat, X search, image/video generation, TTS, and transcription.

Category
Visit Server

README

grok-oauth-mcp

Standalone TypeScript stdio MCP server for xAI Grok using web OAuth PKCE. It is intended for local MCP clients such as Claude Desktop, ChatGPT-compatible MCP hosts, and other stdio MCP launchers.

This server does not require XAI_API_KEY for normal usage. The primary auth path is xAI web OAuth with a local token store.

Requirements

  • Node.js 20+
  • A local MCP client that can launch stdio servers, such as Claude Desktop or another MCP-compatible host.
  • An xAI account with an active SuperGrok subscription or an X Premium+ subscription linked to the X account you sign in with. This matches the Hermes xAI Grok OAuth connector: the browser login works through accounts.x.ai, and xAI links X Premium+ entitlement to the xAI session automatically.

No XAI_API_KEY is used for the normal OAuth path. xAI may still gate OAuth API access by subscription tier; if OAuth succeeds but xAI API calls return 403, your account may not be entitled for that OAuth API surface yet.

Features

  • xAI web OAuth PKCE login with the Hermes client ID.
  • Token storage at ~/.config/grok-oauth-mcp/tokens.json by default.
  • Configurable token directory with GROK_OAUTH_MCP_CONFIG_DIR.
  • Refresh-token flow before access token expiry.
  • Configurable xAI API base with XAI_BASE_URL, defaulting to https://api.x.ai/v1.
  • MCP stdio tools for auth, Grok chat, X search, image, video, TTS, and transcription.
  • Multipart upload support for local image/audio paths.

Quick Start

This is the shortest path for a human or agent to get the MCP running, authenticate, and verify that Grok works.

1. Install

From source:

git clone https://github.com/bowtieswan/grok-oauth-mcp.git
cd grok-oauth-mcp
npm install
npm run build

Run directly from the built checkout:

node /absolute/path/to/grok-oauth-mcp/dist/index.js

Or install it globally from the checkout:

npm install -g .
grok-oauth-mcp

Package-runner form:

npx grok-oauth-mcp

2. Add It To Your MCP Client

Use absolute paths when possible. This avoids PATH differences between your shell and GUI apps.

Claude Desktop

Add an MCP server entry that points at your local build or global executable. Local build example:

{
  "mcpServers": {
    "grok-oauth": {
      "command": "node",
      "args": ["/absolute/path/to/grok-oauth-mcp/dist/index.js"]
    }
  }
}

If you installed globally and know the executable path:

{
  "mcpServers": {
    "grok-oauth": {
      "command": "/absolute/path/to/node",
      "args": ["/absolute/path/to/grok-oauth-mcp"]
    }
  }
}

OpenCode

Add a local MCP server entry under the top-level mcp object in your OpenCode config, usually ~/.config/opencode/opencode.json:

{
  "mcp": {
    "grok-oauth": {
      "command": [
        "/absolute/path/to/node",
        "/absolute/path/to/grok-oauth-mcp"
      ],
      "enabled": true,
      "type": "local"
    }
  }
}

Then restart the MCP client or reload MCP servers. In OpenCode, verify discovery with:

opencode mcp list

You should see grok-oauth connected.

3. Authenticate

Use the MCP tool auth_login with wait=false first. This is the safest mode for most MCP clients because the tool returns the URL immediately.

Input:

{
  "wait": false,
  "timeout_ms": 600000
}

The result contains:

  • auth_url: open this in your browser.
  • state: the OAuth state for this attempt.
  • redirect_uri: normally http://127.0.0.1:56121/callback.
  • token_path: where tokens will be written after success.

Open auth_url, sign in to xAI, then complete one of these two paths.

Path A: Browser Redirects To Localhost

If the browser redirects to http://127.0.0.1:56121/callback?... and shows a success message, login is complete. Run auth_status to confirm.

Path B: xAI Shows A Grok Build Code

Sometimes xAI shows a page saying:

Enter this code to finish signing in
Copy the code below into Grok Build to finish signing in

If that happens, copy the code and call auth_exchange_code immediately:

{
  "code": "paste-the-grok-build-code-here"
}

The code is tied to the most recent auth_login attempt and the saved PKCE verifier. If it expires or you started another login attempt, run auth_login again and use the fresh code from the fresh URL.

If you have a full callback URL instead of a bare code, pass it as callback_url:

{
  "callback_url": "http://127.0.0.1:56121/callback?code=...&state=..."
}

4. Verify Auth

Run auth_status. A successful login returns authenticated: true, has_refresh_token: true, and an expiry time. Token values are never returned.

Tokens are stored at:

~/.config/grok-oauth-mcp/tokens.json

Temporary pending OAuth state is stored at:

~/.config/grok-oauth-mcp/pending_oauth.json

Both files are written with mode 0600 where the local filesystem supports POSIX permissions.

5. Test Grok

After auth_status is authenticated, try a small call. For X search:

{
  "query": "latest xAI Grok updates",
  "filters": {
    "max_results": 5
  }
}

If the first broad search times out, retry with a narrower query and fewer results.

OAuth Details

  • Discovery: https://auth.x.ai/.well-known/openid-configuration
  • Displayed authorization server: accounts.x.ai
  • Client ID: b1a00492-073a-47ea-816f-4c329264a828
  • Scope: openid profile email offline_access grok-cli:access api:access
  • Redirect URI: http://127.0.0.1:56121/callback
  • Authorize URL includes plan=generic and referrer=hermes-agent.

Authentication Notes For Agents

  • Prefer auth_login with wait=false. Do not use wait=true unless your MCP host surfaces stderr or tool output while the call is still running.
  • Always open the exact auth_url from the latest auth_login result.
  • If the page shows a Grok Build code, call auth_exchange_code with that code. Do not try to open /callback manually.
  • If auth_exchange_code returns No pending OAuth login found or Pending OAuth login is expired, run auth_login again and use the fresh URL/code.
  • If xAI returns invalid_grant, the code is stale, already used, or from a different login attempt. Run auth_login again.

Tools

auth_login

Starts OAuth login and returns an authorization URL.

Input:

{
  "wait": false,
  "timeout_ms": 300000
}

Set wait=true only for clients that can show stderr or otherwise surface the URL while the tool is still running.

auth_exchange_code

Exchanges a pasted xAI/Grok Build OAuth code, or a full callback URL, using pending PKCE state from the last auth_login call.

Input with a bare Grok Build code:

{
  "code": "paste-the-grok-build-code-here"
}

Input with a callback URL:

{
  "callback_url": "http://127.0.0.1:56121/callback?code=...&state=..."
}

auth_status

Reports whether a token is stored, whether a refresh token exists, and approximate expiry. Token values are never returned.

auth_logout

Deletes the local token file and any pending OAuth state.

grok_chat

Posts to /responses with store=false. Default model: grok-build-0.1.

Example:

{
  "input": "Explain what this MCP server can do."
}

x_search

Posts to /responses with tools: [{ "type": "x_search" }] and optional filters.

Example:

{
  "query": "latest xAI Grok updates",
  "filters": {
    "max_results": 5
  }
}

grok_tts

Posts to /tts.

Example:

{
  "text": "Hello from Grok.",
  "voice_id": "eve",
  "language": "en"
}

grok_image

Posts to /images/generations for text-to-image and /images/edits when image_path is present.

Defaults:

  • Generation model: grok-imagine-image
  • Edit model: grok-imagine-image-quality

Common image options include aspect_ratio, resolution, size, n, and response_format. Pass values supported by the current xAI/Hermes image surface, such as common aspect ratios like 1:1, 16:9, 9:16, 4:3, and 3:4 when available on your account.

grok_video

Posts to /videos/generations. Supports text-to-video and image-to-video when image_path is present.

Defaults:

  • Text-to-video model: grok-imagine-video
  • Image-to-video model: grok-imagine-video-1.5-preview

Set poll=true to poll /videos/{request_id} until the request leaves a queued or processing state.

grok_transcribe

Posts a local audio file to /stt with multipart form data, matching xAI Grok STT.

Example:

{
  "file_path": "/path/to/audio.wav",
  "language": "en"
}

Troubleshooting

  • Not authenticated. Run auth_login first.: Run auth_login with wait=false, open the returned auth_url, then complete the redirect or use auth_exchange_code with the Grok Build code.
  • Browser cannot reach callback: If xAI showed a Grok Build code, do not manually open /callback; call auth_exchange_code with the code instead. If you expected a redirect, make sure nothing else is using port 56121, then run auth_login again.
  • No pending OAuth login found or Pending OAuth login is expired: Run auth_login again and use the fresh URL/code.
  • invalid_grant: The pasted code is stale, already used, or belongs to a different login attempt. Run auth_login again and paste the fresh code.
  • 403 from xAI: Your xAI account may not have the required SuperGrok or X Premium+ entitlement for the requested Grok API surface.
  • Token refresh fails: Run auth_logout, then auth_login again.
  • Custom API gateway or mock server: Set XAI_BASE_URL to the replacement base URL.
  • Custom token location: Set GROK_OAUTH_MCP_CONFIG_DIR to a directory path. The server writes tokens.json inside it.

Development

npm test
npm run build

Tests mock all OAuth and xAI network calls. They must not call real xAI OAuth or xAI APIs.

License

MIT

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured