MCP host 的鉴权 + 端点布局见 MCP 概览。反向 ——
智能体通过
attachments[] 接收文件、以及跨 OpenAPI 和 A2A 的 file_id
模型 —— 见 工具 § 附件 和
调用智能体 § 附件。资源
资源代表 MCP 客户端可读不可改的数据。resources/list
发现可用资源：{
"jsonrpc": "2.0",
"id": 1,
"method": "resources/list"
}
{
"jsonrpc": "2.0",
"id": 1,
"result": {
"resources": [
{
"uri": "beeos://agent/config",
"name": "Agent Configuration",
"description": "Current agent configuration and capabilities",
"mimeType": "application/json"
},
{
"uri": "beeos://agent/skills",
"name": "Skill Registry",
"description": "List of all registered skills with metadata",
"mimeType": "application/json"
}
]
}
}
资源 schema
| 字段 | 类型 | 说明 |
|---|---|---|
uri | string | 资源唯一标识 |
name | string | 人类可读名 |
description | string | 资源包含啥 |
mimeType | string | 资源 content type |
resources/read
按 URI 读特定资源：{
"jsonrpc": "2.0",
"id": 2,
"method": "resources/read",
"params": {
"uri": "beeos://agent/config"
}
}
{
"jsonrpc": "2.0",
"id": 2,
"result": {
"contents": [
{
"uri": "beeos://agent/config",
"mimeType": "application/json",
"text": "{\"name\":\"Research Assistant\",\"model\":\"gpt-4o\",\"skills\":[\"web_search\",\"summarize\"]}"
}
]
}
}
Prompt 模板
Prompt 模板是 MCP 客户端可呈现给用户的可复用消息模式。prompts/list
{
"jsonrpc": "2.0",
"id": 3,
"method": "prompts/list"
}
{
"jsonrpc": "2.0",
"id": 3,
"result": {
"prompts": [
{
"name": "research",
"description": "Research a topic and provide a summary",
"arguments": [
{
"name": "topic",
"description": "The topic to research",
"required": true
},
{
"name": "depth",
"description": "Research depth: brief, standard, or deep",
"required": false
}
]
}
]
}
}
prompts/get
按参数渲染 prompt 模板：{
"jsonrpc": "2.0",
"id": 4,
"method": "prompts/get",
"params": {
"name": "research",
"arguments": {
"topic": "quantum computing",
"depth": "deep"
}
}
}
{
"jsonrpc": "2.0",
"id": 4,
"result": {
"messages": [
{
"role": "user",
"content": {
"type": "text",
"text": "Research quantum computing in depth. Provide a comprehensive summary covering recent advances, key players, and practical applications."
}
}
]
}
}
示例：完整资源流程
# 列资源
curl -s -X POST "https://mcp.beeos.ai/${AGENT_ID}/mcp" \
-H "X-Agent-API-Key: bak_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"resources/list"}' | jq '.result.resources'
# 读一个资源
curl -s -X POST "https://mcp.beeos.ai/${AGENT_ID}/mcp" \
-H "X-Agent-API-Key: bak_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":2,"method":"resources/read","params":{"uri":"beeos://agent/config"}}' | jq
# 列 prompts
curl -s -X POST "https://mcp.beeos.ai/${AGENT_ID}/mcp" \
-H "X-Agent-API-Key: bak_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":3,"method":"prompts/list"}' | jq '.result.prompts'
可用性
不是所有智能体都暴露资源或 prompt。智能体没注册过的话响应可能是空数组：{
"jsonrpc": "2.0",
"id": 1,
"result": {
"resources": []
}
}