Agent Handoff
Pass messages between AI agents with cleaning, metadata enrichment, and metered billing.
我该使用它吗
质量与安全性
A
基于对工具定义和协议合规性的自动分析。
上下文开销
~209token 数（工具定义）
~566 B典型响应大小
对注意力的影响极小（占 128k 上下文窗口的 0.16%）
这是每次将服务器的工具加载到模型上下文窗口时所消耗的大致 token 数。数值越高，可用于其他任务的注意力就越少。
安装
一键安装
将以下内容添加到你的 `claude_desktop_config.json` 文件中：
{
"mcpServers": {
"agent-handoff": {
"url": "https://agent-handoff-production-573c.up.railway.app/mcp"
}
}
}远程端点
https://agent-handoff-production-573c.up.railway.app/mcpstreamable-http它能做什么
工具清单
工具（2）
🟢 只读🟡 写入🔴 删除⚪ 未知
⚪handoff(message, next_agent)
Hand a message off to another agent. Cleans and normalizes the payload, stamps it with handoff metadata (id, timestamp, estimated tokens, target agent), and records the handoff. Use this when passing work or context from one agent to the next in a multi-agent pipeline. This is a metered, billable operation.
输入模式
{
"type": "object",
"properties": {
"message": {
"description": "The payload to hand off. Accepts a plain string or a JSON object. Max 256 KB.",
"type": [
"string",
"object",
"array",
"number",
"boolean"
]
},
"next_agent": {
"type": "string",
"description": "Name/identifier of the agent receiving this handoff.",
"maxLength": 128
}
},
"required": [
"message",
"next_agent"
]
}🟢get_usage_stats
Return this account's total handoff count and the most recent handoffs. Read-only and not billed.
输入模式
{
"type": "object",
"properties": {}
}社区
证据
最近观测
已验证未记录版本2 个工具
已验证未记录版本2 个工具
已验证未记录版本2 个工具