Protocol Specification

Model Context Protocol (MCP v2.4)

AgentBoost implements the open Model Context Protocol over HTTP/JSON-RPC 2.0. Any external host—including Claude Desktop, Cursor, and custom enterprise workers—can discover and execute tools deterministically.

Phase 1 • Connection

Client Handshake (initialize)

Upon connection, external AI clients dispatch an initialize method to negotiate protocol capabilities. AgentBoost responds with supported tool schemas and server identity.

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "initialize",
  "params": {
    "protocolVersion": "2024-11-05",
    "capabilities": { "tools": {} },
    "clientInfo": { "name": "claude-desktop", "version": "1.0.0" }
  }
}
Phase 2 • Reflection

Tool Discovery (tools/list)

Hosts request the available capability catalog. AgentBoost exposes all 60+ verified skills formatted with standard JSON Schema input constraints.

{
  "jsonrpc": "2.0",
  "id": 2,
  "result": {
    "tools": [
      {
        "name": "crypto_quant_pro",
        "description": "Calculates options Greeks and volatility hedges.",
        "inputSchema": {
          "type": "object",
          "properties": {
            "asset": { "type": "string", "description": "Ticker symbol (BTC, ETH, SOL)" },
            "timeframe": { "type": "string" }
          },
          "required": ["asset"]
        }
      }
    ]
  }
}
Phase 3 • Execution

Tool Invocation (tools/call)

When an AI model needs deterministic calculations (e.g., calculating Cap Rates, checking idle EC2 compute, scoring B2B leads), it dispatches a tools/call request to the AgentBoost endpoint.

{
  "jsonrpc": "2.0",
  "id": 3,
  "method": "tools/call",
  "params": {
    "name": "crypto_quant_pro",
    "arguments": {
      "asset": "BTC",
      "timeframe": "7D"
    }
  }
}

Ready to connect Claude Desktop to AgentBoost?

Generate an API key in your workspace dashboard to start running tools in Claude.

Get Access Key