🤖 MCP Client - AI Bridge

Bridge AI models (Claude, GPT) with MCP servers for dynamic tool discovery and execution

Server Running

⏱️ Uptime: 1h 55m

📊 Version: 1.0.0

🔗 MCP Server: Configured

📡 API Endpoints

GET /

This documentation page

POST /ask

Main endpoint - Send prompts to AI models with dynamic MCP tool discovery

prompt The question or request for the AI model (required)

apiKey Your API key for the selected model (required)

model "anthropic" or "openai" (default: "anthropic")

maxRounds Maximum tool use rounds (default: 10)

GET /health

Health check with system metrics

{
  "status": "healthy",
  "timestamp": "2026-01-19T05:45:54.755Z",
  "uptime": 6928,
  "memory": {
    "used": 11,
    "total": 18,
    "percentage": 61
  }
}

GET /health/ready

Readiness check including MCP server connectivity

🤖 Supported AI Models

Anthropic Claude

Access Claude 3 models with advanced reasoning and tool use capabilities

claude-3-opus claude-3-sonnet claude-3-haiku

OpenAI GPT

Use GPT-4 and GPT-3.5 models with function calling

gpt-4-turbo gpt-4 gpt-3.5-turbo

💡 Examples

Using Claude 3

{
  "prompt": "What astronomical observations are available for NGC 1234?",
  "apiKey": "sk-ant-...",
  "model": "anthropic"
}

Using GPT-4

{
  "prompt": "List the available CADC tables and their purposes",
  "apiKey": "sk-...",
  "model": "openai"
}

Automatic Tool Discovery

{
  "prompt": "Query the top 5 observations from CFHT and resolve their DataLinks",
  "apiKey": "your-api-key",
  "model": "anthropic",
  "maxRounds": 5
}

// The client will automatically:
// 1. Connect to the MCP server
// 2. Discover available tools (query_adql, resolve_datalink)
// 3. Let the AI model use these tools
// 4. Return the final response

Response Format

{
  "response": "Based on the CADC database, I found 5 observations from CFHT...",
  "toolsUsed": ["query_adql", "resolve_datalink"],
  "rounds": 2
}

🔧 How It Works

1. Dynamic Tool Discovery

On each request, the client connects to the MCP server and discovers available tools

2. AI Model Integration

Tools are converted to the format expected by your chosen AI model (Claude or GPT)

3. Automatic Tool Execution

When the AI requests tool use, the client executes it via MCP and returns results

4. Iterative Processing

The process continues until the AI has a final answer or reaches the maximum rounds

🔐 Security

API Key Handling

• API keys are provided per-request and never stored
• Keys are passed directly to the AI provider
• No server-side credential storage
• Secure HTTPS communication required

📊 Configuration

Environment Variables

MCP_SERVER_URL URL of your MCP server (default: http://localhost:4000/mcp)

PORT Port to run the client on (default: 3000)