
Fleetpost
io.github.StanimirTenev由 GitHub 账号 StanimirTenev 发布v0.2.0更新于 Oct 25, 2026
A mailbox for your machines: read fleet capabilities, send a request. Offline, no server.
概览
Fleetpost 是一个本地 MCP 服务器,为助手提供“机器邮箱”,通过共享的 rclone 远端在多台机器之间协调任务,无需中心服务器。
- 可以做什么
- 查看机群状态:待处理请求、协议变更、已知机器和上次周期,以及各机器能力,均读取已同步文件,可离线使用。从本机收件箱读取单条请求。向另一台机器的收件箱发送请求,必须提供需求、完成标准和过期时间。将请求标记为已处理,移入本机收件箱的已处理目录。立即执行一次协调周期,这是少数会访问远端的操作之一。
- 适用场景
- 适用于跨多台并非始终在线的机器协调工作:让助手查询各机器能做什么、分派任务,并通过共享文件夹而非服务器或守护进程处理收到的请求。
- 运行要求
- 以 stdio 本地运行,通常用 uvx 从 fleetpost-mcp 的 PyPI 包启动。需要可用的 Fleetpost 代码检出及填写好的 config.env,通过 FLEETPOST_CONFIG 指定,或放在启动目录中。若配置被复制到检出目录之外,需用 FLEETPOST_HOME 指向检出目录,因为 fleet_sync 和 fleet_handle 会运行仓库自带脚本。发送和同步需要 rclone 远端。
安装
在 SourceWeft 中
- 打开 控制台中的 Fleetpost,将其添加到工作区。
- 为需要使用其工具的对话启用该服务。
Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。
其他 MCP 客户端
参照 仓库 中的启动说明。
README
fleetpost-mcp
Give your agent the fleet's mailbox. An MCP server for Fleetpost — offline-tolerant, server-less coordination between machines over any rclone remote.
Your agent can ask what the other machines can do, hand one of them a task, and work through its own inbox — without a server, a daemon, or every machine being online at once. A machine that is off picks up its mail on its next cycle.
Tools
Only fleet_send_request and fleet_sync touch the remote. Everything else reads files the
sync already pulled, so status is instant and works with no network at all.
Setup
You need a working Fleetpost checkout with a filled-in config.env — see the
main README. The MCP server reads that same
file; it introduces no configuration of its own.
FLEETPOST_CONFIG can be omitted if the server starts in the directory holding config.env.
If you copied the config away from the checkout, also set FLEETPOST_HOME to the checkout —
fleet_sync and fleet_handle run the repo's own scripts.
Then ask your agent:
Anything waiting for me? — and if the build box can sign, ask it to sign the release.
What it will not do
The protocol's rules are enforced here, not just documented:
- Writes only into another machine's inbox, at the top level. Never into
handled/, never into anyone's descriptors. That is what keeps concurrent writes safe. fleet_handleonly tidies your own inbox — the owner is the only one allowed to.- A request must be complete to be sent. What is wanted, what "done" means, and an expiry are required arguments; a malformed request cannot be created.
- Ask for actions, not access. If a job needs a credential you lack, the machine that has it does the work and returns the result. Secrets never travel.
There is no server, no database, and no daemon of our own — the transport stays a folder on an rclone remote, and this server only reads and writes files in it.
License
MIT.
来源:mcp/README.md,提交 d2e4f9c
工具
0版本历史
1- v0.2.0最新Sep 16, 2026
评分
已满足适用检查项的 50%
该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。
- 近 30 天内可连通25 分
请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。
- 已声明认证方式10 分
请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。
- 已列出工具15 分
请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。
- 描述必需15 分
- 图标10 分
- 文档15 分
- 许可证10 分