mcp-server-grok-chat

An MCP (Model Context Protocol) server for the xAI Grok API. Built in Rust, exposes chat completions, vision, web/X search, embeddings, and model listing as MCP tools.

Communicates via stdio using JSON-RPC 2.0, like all MCP servers.

Tools

Tool	Description
`chat`	Send a chat completion request to Grok with optional multi-turn history, system prompt, structured output (JSON schema), model selection, and multi-agent research
`chat_with_vision`	Analyse an image with Grok's vision capabilities given an image URL and text prompt
`chat_with_search`	Chat with Grok using live web search and/or X (Twitter) search to ground responses
`embedding`	Generate text embeddings using Grok's embedding model
`list_models`	List all available Grok models and their IDs (cached for 5 minutes)

chat

Send a chat completion request. Supports multi-turn conversations via a JSON message history array, system prompts, structured output via JSON schema, temperature control, model selection, and multi-agent research.

When using a multi-agent model (any model ID containing multi-agent), the request is automatically routed through the Responses API. The multi-agent model dispatches your query to multiple agents that research in parallel, then synthesizes their findings. Use reasoning_effort to control agent count. Call the list_models tool to see which multi-agent models are currently available.

Parameters:

Name	Type	Required	Description
`prompt`	string	yes	The user message to send
`model`	string	no	Model ID (default: `grok-4.3`). Call `list_models` for the current set.
`system_prompt`	string	no	System prompt to set context
`messages`	string	no	Full conversation history as JSON array of `{role, content}` objects
`temperature`	float	no	Sampling temperature (0.0 - 2.0)
`max_tokens`	integer	no	Maximum tokens to generate
`response_schema`	string	no	JSON schema string to enforce structured output
`reasoning_effort`	string	no	On `grok-4.3`: `low`/`medium`/`high` controls native reasoning depth. On multi-agent models: `low`/`medium` = 4 agents, `high`/`xhigh` = 16 agents (`xhigh` is multi-agent-only).

chat_with_vision

Analyse an image using Grok's vision capabilities.

Parameters:

Name	Type	Required	Description
`prompt`	string	yes	Text prompt describing what to analyse
`image_url`	string	yes	URL of the image (must be http:// or https://)
`model`	string	no	Model ID (default: `grok-4.3`). Must be a vision-capable model. Call `list_models` for the current set.
`detail`	string	no	Image detail level: `low` or `high` (default: `high`)
`temperature`	float	no	Sampling temperature (0.0 - 2.0)
`max_tokens`	integer	no	Maximum tokens to generate

chat_with_search

Chat with Grok using live web search and/or X (Twitter) search. The model automatically searches the internet to ground its response.

Parameters:

Name	Type	Required	Description
`prompt`	string	yes	The user message to send
`search_type`	string	no	Search type: `web`, `x`, or `both` (default: `both`)
`model`	string	no	Model ID (default: `grok-4.3`). Call `list_models` for the current set.
`system_prompt`	string	no	System prompt to set context
`temperature`	float	no	Sampling temperature (0.0 - 2.0)
`max_tokens`	integer	no	Maximum tokens to generate
`reasoning_effort`	string	no	On `grok-4.3`: `low`/`medium`/`high` controls native reasoning depth. On multi-agent models: `low`/`medium` = 4 agents, `high`/`xhigh` = 16 agents (`xhigh` is multi-agent-only).

embedding

Generate text embeddings.

Parameters:

Name	Type	Required	Description
`input`	string	yes	Text to embed as JSON: a single string or array of strings
`model`	string	no	Embedding model to use (default: `grok-2-text-embedding`)

list_models

List all available Grok models. No parameters. Results are cached for 5 minutes.

Prerequisites

Rust (edition 2024)
An xAI API key from console.x.ai

Setup

Create the config file:

bash

mkdir -p ~/.config/mcp-server-grok-chat

Create ~/.config/mcp-server-grok-chat/config.toml:

toml

api_key = "xai-..."

Build

bash

cargo build --release

This produces target/release/grok-chat.

For development:

bash

cargo build              # debug build
cargo run                # run in dev mode
RUST_LOG=debug cargo run # run with debug logging

MCP Configuration

Add to your Claude Desktop config (~/.config/Claude/claude_desktop_config.json):

json

{
  "mcpServers": {
    "grok-chat": {
      "command": "/path/to/grok-chat"
    }
  }
}

Project Structure

src/
  main.rs    - entry point, config loading, stdio transport setup
  server.rs  - MCP tool definitions (chat, chat_with_vision, chat_with_search, embedding, list_models)
  api.rs     - xAI HTTP client, request/response types, response formatters
  params.rs  - tool parameter types with serde and JSON Schema derives
  config.rs  - TOML config loading