MCP · Model Context Protocol · V5.0

黄庭协议 MCP 服务器

将 Agent 工作流成本优化能力封装为标准 MCP 工具。一次调用，三阶段降本：压缩输入、剪枝过程、精炼输出。

MCP 端点https://mcp.huangting.ai/mcp

无需认证 · 无需 API Key · 开放接入

快速接入

1. Claude Desktop / Cursor

在配置文件中添加以下内容，重启后 Agent 将自动发现 create_optimization_context 工具：

{
  "mcpServers": {
    "huangting-flux": {
      "url": "https://mcp.huangting.ai/mcp",
      "transport": "http"
      // No authentication required — open access
    }
  }
}

配置文件位置：macOS ~/Library/Application Support/Claude/claude_desktop_config.json

2. Manus Agent

在 Manus 设置 → MCP 服务器中添加，Agent 将在每个任务开始时自动调用 create_optimization_context：

{
  "mcpServers": {
    "huangting-flux": {
      "url": "https://mcp.huangting.ai/mcp"
      // Manus auto-handles OAuth discovery — no manual auth needed
    }
  }
}

3. 直接调用 JSON-RPC（Python 示例）

通过标准 HTTP POST 调用 MCP 端点，无需安装任何 SDK：

import requests, json

# Step 1: call at task start — get three-stage optimization plan
resp = requests.post("https://mcp.huangting.ai/mcp", json={
    "jsonrpc": "2.0", "id": 1,
    "method": "tools/call",
    "params": {
        "name": "create_optimization_context",
        "arguments": {"task_description": "Analyze Q1 sales data and write a report"}
    }
})
plan = json.loads(resp.json()["result"]["content"][0]["text"])

# Step 2: use the compressed core instruction as your task prompt
core = plan["stages"][0]["payload"]["core_instruction"]
context_id = plan["context_id"]
baseline = plan["baseline_estimate"]["total_tokens"]

# ... run your task using core as the guiding instruction ...

# Step 3: report actual token usage after task completes
requests.post("https://mcp.huangting.ai/mcp", json={
    "jsonrpc": "2.0", "id": 2,
    "method": "tools/call",
    "params": {
        "name": "report_optimization_result",
        "arguments": {
            "agent_id": "my-agent-001",
            "context_id": context_id,
            "actual_tokens_used": 3200,
            "baseline_tokens": baseline
        }
    }
})

curl 快速测试：

curl -X POST https://mcp.huangting.ai/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "method": "tools/call",
    "params": {
      "name": "create_optimization_context",
      "arguments": { "task_description": "Write a market analysis report" }
    },
    "id": 1
  }'