GPU-Bridge

Posted on Mar 18

MCP for AI Services: How to Give Claude Desktop Access to 30 GPU-Powered Tools

#programming #ai #tutorial #productivity

Claude Desktop can browse the web. It can read files. But can it generate images, transcribe audio, or run LLM inference on open-source models?

With MCP (Model Context Protocol) and GPU-Bridge, yes — in about 30 seconds.

What Is MCP?

MCP is an open protocol (created by Anthropic) that lets AI models use external tools. Think of it as a plugin system for LLMs:

Claude Desktop ← MCP Protocol → Tool Server → External Service

Any MCP-compatible tool server can be plugged into Claude Desktop, Cursor, Windsurf, or any MCP client. The model discovers available tools and uses them as needed.

Setting Up GPU-Bridge MCP

Step 1: Get an API Key

Step 2: Configure Claude Desktop

Add this to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on Mac):

{
  "mcpServers": {
    "gpu-bridge": {
      "command": "npx",
      "args": ["-y", "@gpu-bridge/mcp-server"],
      "env": {
        "GPUBRIDGE_API_KEY": "your-api-key-here"
      }
    }
  }
}

Step 3: Restart Claude Desktop

That's it. Claude now has access to 30 AI services.

What Can You Do?

Once connected, Claude can use these tools:

🎨 Image Generation

"Generate an image of a futuristic Tokyo street at night"

Claude calls gpu_bridge_run with service: "image-sdxl" and returns the generated image.

🔤 Text Embeddings

"Create embeddings for these 100 product descriptions and find the most similar pairs"

Claude calls the embeddings service, gets vectors, and computes similarity — all within the conversation.

🗣️ Speech to Text

"Transcribe this audio file"

Claude uses the transcription service to convert speech to text.

📄 Document Parsing

"Extract all the text and tables from this PDF"

Claude calls the document parser and returns structured content.

🤖 Open-Source LLMs

"Ask Llama 3.3 70B to review this code"

Claude routes the request to Groq's Llama inference and returns the response. Yes, Claude can delegate to other LLMs for specialized tasks.

The 5 MCP Tools

GPU-Bridge exposes 5 MCP tools:

Tool	Description
`gpu_bridge_run`	Execute any of 30 AI services
`gpu_bridge_services`	List available services with pricing
`gpu_bridge_models`	Get models available for a service
`gpu_bridge_health`	Check API status
`gpu_bridge_docs`	Get usage documentation

The gpu_bridge_run tool is the workhorse. It accepts a service name and input, routes to the right GPU provider, and returns the result.

Real Workflow Example

Here's a realistic use case — building a research assistant:

You: "Read this research paper PDF, extract the key findings, generate embeddings for each finding, and create a summary image that visualizes the main concepts."

What Claude does:

Calls gpu_bridge_run with service: "document-parse" → extracts text from PDF
Processes the text to identify key findings
Calls gpu_bridge_run with service: "embeddings" → generates vectors for semantic clustering
Groups findings by similarity
Calls gpu_bridge_run with service: "image-sdxl" → generates a concept visualization
Presents everything in a coherent summary

Four GPU-powered operations in one conversation. No switching apps, no managing APIs.

Pricing

MCP tools are billed per-use through your GPU-Bridge account:

Operation	Approximate Cost
Image generation	$0.003-0.005
1K token embedding	$0.00003
Document parsing	$0.002
LLM inference (1K tokens)	$0.0006-0.003

A typical research session with 20 tool calls might cost $0.05-0.10.

Beyond Claude Desktop

GPU-Bridge MCP works with any MCP-compatible client:

Cursor — AI coding with GPU-powered tools
Windsurf — Same setup, different editor
Custom agents — Any MCP client library

The MCP server is also available as a hosted HTTP endpoint:

POST https://api.gpubridge.io/mcp

This means even web-based agents can use it without running a local server.

Getting Started

# Try it immediately (no install)
npx @gpu-bridge/mcp-server

# Or install globally
npm install -g @gpu-bridge/mcp-server

The npm package is @gpu-bridge/mcp-server — currently at v2.4.3.

What would you build with 30 AI services inside Claude Desktop? Drop your ideas — I'm curious what use cases people come up with.

DEV Community