Swarm Rd Orchestrator

MCP文档知识

提供方 未标注

文档知识 未标注 未标注 社区 / 第三方 Claude DesktopClaudeCursorVS Code

仅追加事件日志,用于通过3个MCP工具实现跨代理的上下文/记忆共享。

一句话选型结论

按资料字段整理

Swarm Rd Orchestrator 属于文档知识类 MCP 服务器,传输方式未标注,鉴权方式未标注,可直接接入 Claude Desktop、Claude、Cursor 等,没有现成安装命令,需按官方文档接入。接入前先确认权限范围;涉及写入、删除或付款的操作,建议先在测试环境验证。

分类文档知识 传输方式未标注 鉴权未标注 部署方式未标注
GitHub 星标 未标注 仓库热度,无公开仓库时显示未标注
工具数 未标注 暴露给模型的工具数量
资源 / 提示词 0 / 0 资源与提示词模板数量
维护方 社区 / 第三方 核验等级 仅资料核对

工具与能力

文档知识
数量来自来源目录的标注,具体工具清单以官方文档为准
工具数量未标注暴露给模型可调用的工具个数
资源数量0可读取的资源条目
提示词数量0内置提示词模板数量

能力标签

支持客户端

适用场景

记忆

安装与配置

未标注 未标注
上游未给出传输方式
  1. 1
    准备运行环境

    按官方文档确认运行环境(常见为 Node.js 或 Python),装好后再执行安装命令。

  2. 2
    安装服务

    复制上面的安装命令执行;命令里带占位符的位置,换成自己的密钥或路径。

  3. 3
    写入客户端配置

    在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,让新配置生效。

  4. 4
    验证是否可用

    连接成功后随便调一个只读工具,确认能返回结果,再接入正式流程。

典型使用场景

3 类
按分类与标签整理的常见用法,具体以官方文档为准
文档检索问答

在你的文档库里按问题找答案。

知识库读写

把新资料写进知识库,随用随取。

内容整理

把零散资料归纳成结构化笔记。

选择建议

选之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 已适配 Claude Desktop、Claude、Cursor,换客户端不用重新研究配置。

先想清楚的地方

  • 没有现成安装命令,需要按官方文档自己填启动命令和参数。
  • 资料没标注工具数量,实际能力要装好后确认。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在文档知识场景里让模型直接调用外部能力的人。
  • 已经在用 Claude Desktop、Claude 的用户,接入路径最短。
  • 想先快速试一下、能接受在本地或自有环境跑服务的人。

不太适合

  • 需要厂商级 SLA 和技术支持的生产环境。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
MCP 服务器工具数传输方式部署方式安装命令最后核验
Swarm Rd Orchestrator当前 未标注 未标注 未标注 无 未标注
dify 未标注 stdio · 本地进程 本地部署 有 2026-04-11
open-webui 未标注 stdio · 本地进程 本地部署 有 2026-04-11
JeecgBoot 未标注 stdio · 本地进程 本地部署 有 2026-04-11
Cherry Studio 未标注 未标注 未标注 无 2026-05-17
mempalace 未标注 stdio · 本地进程 本地部署 有 2026-04-11

服务器档案

文档知识
字段来自项目仓库与公开索引,中文说明由本站标注
提供方未标注上游标注的维护方 未标注
分类文档知识本站按上游分类展示
连接方式未标注上游未给出传输方式
鉴权方式未标注远程服务建议用最小权限凭据
维护方社区 / 第三方核验等级 仅资料核对
包名 / 运行时未标注 · 未标注本地安装时用到的包标识与运行时
部署类型未标注上游给出的部署形态
最近核验未标注上游最后核验时间

详细介绍

10.5 KB
来自项目仓库与上游目录的原文,未做删改

swarm-rd-orchestrator-cli

Ray-native context and memory sharing for parallel research agents, with an agent-native CLI and MCP server.

An append-only, SQLite-WAL-backed event log wrapped as a Ray actor, so parallel agents can write findings and pull each other's without a shared mutable store, plus a CLI and MCP server so both humans and other agents can drive it directly.

This is a Milestone 1 prototype (2026-08-03): validate the approach on a real task before building further. See Locked decisions below and spike.py for the actual validation harness.

Install```bash

pip install swarm-rd-orchestrator-cli

or

npm install -g swarm-rd-orchestrator-cli


Either gives you a `swarm-rd-cli` command on your `PATH`. The npm package is a thin wrapper around the Python CLI: it execs the real binary, it does not reimplement it. Install the Python package too if you use the npm one.

**Status:** live on both registries (PyPI published via GitHub Actions OIDC, no stored token). Both were verified with a real install and a real command run in a clean environment, not just a successful upload.

> [!WARNING]
> This is a pre-validation Milestone 1 spike (`0.0.x`), not yet a stable release. Tested on macOS only; Windows support is unverified since Ray's own Windows support is more limited than Linux/macOS upstream.

## Quickstart```bash
swarm-rd-cli append task-1 agent-a "found a race condition in the retry loop" --kind result
swarm-rd-cli append task-1 agent-b "confirmed: retry loop isn't holding the lock" --kind result
swarm-rd-cli pull task-1
# [1] (agent-a/result) found a race condition in the retry loop
# [2] (agent-b/result) confirmed: retry loop isn't holding the lock

swarm-rd-cli --json pull task-1     # structured output for scripts/agents
swarm-rd-cli list-tasks             # every task_id with a delta count
swarm-rd-cli mcp                    # run as an MCP server over stdio

Every data-returning command supports --json for agent and script consumption, no screen-scraping required. The mcp subcommand exposes append_delta, pull_deltas, and list_tasks as typed MCP tools over stdio, so an agent can call this programmatically instead of shelling out.

Features

  • Atomic append, proven, not just claimed. A dedicated test simulates a crash mid-write and confirms the SQLite WAL layer leaves zero partial rows, not just a description of the guarantee.
  • Structured output on every data command. --json on append, pull, and list-tasks means an agent shelling out to this CLI never has to screen-scrape human-formatted text.
  • An MCP server, not just a CLI. swarm-rd-cli mcp exposes the same three operations as typed tools over stdio, so an agent can call this programmatically instead of spawning a subprocess.
  • Concurrency tested with real Ray actors, not mocked. The load-bearing test runs 3 actual Ray actors appending concurrently and reconciles the result, the same mechanism the real workload uses.
  • Malformed input fails loudly. A delta missing task_id, agent_id, or content raises InvalidDeltaError before touching storage. No silent drops.

Command reference

Generated from the CLI's own --help output:```
usage: swarm-rd-cli [-h] [--db DB] [--json] {append,pull,list-tasks,mcp} ...

positional arguments:
{append,pull,list-tasks,mcp}
append append a delta to the event log
pull pull deltas for a task
list-tasks list every task_id with a delta count
mcp run as an MCP server over stdio

options:
-h, --help show this help message and exit
--db DB path to the event log (default: swarm-events.db)
--json structured JSON output (for agent/script use)

usage: swarm-rd-cli append [-h] [--kind {note,result,tool_output}]
                            task_id agent_id content

positional arguments:
  task_id
  agent_id
  content

options:
  -h, --help            show this help message and exit
  --kind {note,result,tool_output}

usage: swarm-rd-cli pull [-h] [--since SINCE] task_id

positional arguments:
task_id

options:
-h, --help show this help message and exit
--since SINCE cursor to pull after


## MCP Server

`swarm-rd-orchestrator-cli` ships a Model Context Protocol (MCP) server, so an agent can call the event log directly as typed tools over stdio instead of shelling out to the CLI and parsing text.```bash
pip install "swarm-rd-orchestrator-cli[mcp]"

Run it with:```bash
swarm-rd-cli mcp


Add it to Claude Desktop (or any other MCP client) by pointing it at that command in your config:```json
{
  "mcpServers": {
    "swarm-rd-orchestrator": {
      "command": "swarm-rd-cli",
      "args": ["mcp"]
    }
  }
}

Three tools are exposed:

  • append_delta(task_id, agent_id, content, kind="note") - append a finding/result/tool-output to the event log for a task. Example: append_delta(task_id="task-1", agent_id="agent-a", content="found a race condition in the retry loop", kind="result").
  • pull_deltas(task_id, since_cursor=0) - pull every delta for a task with id greater than since_cursor, oldest first. Example: pull_deltas(task_id="task-1") returns the full history; pull_deltas(task_id="task-1", since_cursor=2) returns only deltas written after cursor 2.
  • list_tasks() - list every task_id currently in the event log with its delta count. Example: list_tasks() returns [{"task_id": "task-1", "delta_count": 2}].

Comparison

swarmmesh is a sibling project in this author's portfolio, also published as swarmmesh-cli on PyPI and npm. It's the more complete option today on almost every dimension below. This project exists as a deliberately Ray-native alternative, not because swarmmesh falls short.

swarm-rd-orchestrator-cli swarmmesh-cli
Transport Ray actor (in-process / distributed) HTTP server
Storage SQLite, WAL mode In-memory by default, or SQLite via --persist
Cross-language Python only Python and Node
Memory search/ranking None (pull by task_id only) BM25 keyword ranking on memory queries
MCP server Yes Yes
Published on PyPI/npm Yes, live Yes, live
CI Yes Yes

If the Ray-native distributed-compute angle doesn't end up mattering for your use case, use swarmmesh instead. It's live, tested against real usage, and does more.

What is swarm-rd-orchestrator-cli, and why does it exist

swarm-rd-orchestrator-cli is a shared, durable event log for parallel AI research agents built on Ray's actor model. Each agent writes findings as structured deltas; any other agent can pull the full history for a task without a shared mutable store or a coordinating server process.

It exists to test a specific, narrow hypothesis: that Ray's actor and object-store model is a better fit for coordinating genuinely large numbers of parallel research agents than an HTTP-based coordination layer. That hypothesis is unproven. The project ships as a Milestone 1 spike specifically to test it against a real workload before any further investment, see Run the actual validation spike.

Build from source```bash

git clone https://github.com/RudrenduPaul/swarm-rd-orchestrator.git
cd swarm-rd-orchestrator
python3 -m venv .venv
.venv/bin/pip install -e ".[dev,mcp]"


Requires Python 3.10 or newer (needed for the `mcp` SDK dependency).

## Run the tests```bash
.venv/bin/pytest test_event_log.py -v

9/9 passing, including the load-bearing test: 3 concurrent Ray actors appending to one shared event log, reconciled with zero lost or duplicated deltas.

Run the actual validation spike

Edit REAL_TASK_ID, REAL_TASK_DESCRIPTION, and the sample agent findings in spike.py to reflect a real research task, then:bash .venv/bin/python3 spike.py

Read the printed rubric at the end. The decision rule: fewer than 3 qualifying architectural failure cases against raw Ray or LangGraph means falling back to a thin CLI wrapper instead of building this out further.

Locked decisions (2026-08-03)

  • Primitive: Ray (actor model and object store), LangGraph as fallback
  • Storage: SQLite, WAL mode
  • Delta shape: {task_id, agent_id, timestamp, content, kind}
  • Malformed delta raises InvalidDeltaError, never silent
  • License: Apache 2.0
  • Python: 3.10 or newer, required for the mcp SDK
  • Publishing: live on PyPI (via GitHub Actions OIDC Trusted Publishing, no stored token) and npm, both verified with a real clean-environment install

FAQ

What does this actually do?
It gives parallel AI agents, running as Ray actors, a shared and durable place to write findings and pull each other's, without stepping on each other's state. It's a transport and persistence layer, not an orchestration framework: it doesn't schedule agents or decide what they do.

How is this different from swarmmesh?
See Comparison above. Same core idea, different transport (Ray actors instead of HTTP), and swarmmesh is currently the more complete, already-published option.

Does this work on Windows, macOS, and Linux?
Tested on macOS. Ray itself supports Linux and macOS natively; Windows support for Ray is more limited upstream, so treat Windows as unverified for this project specifically until someone confirms it.

Does this need my own API keys?
No. This project makes no LLM calls of its own. It's a coordination layer that your own agents, whatever model or framework they use, write to and read from.

Is this safe to depend on?
No, not yet. This is a pre-validation Milestone 1 spike (versions 0.0.x), not a stable release. The event log's core guarantees (atomic append, no partial writes) are tested in test_event_log.py, but the project hasn't been validated against a real multi-agent workload beyond the spike script yet.

How do I use this from an agent, not just a human?
Either shell out to the CLI with --json on every command, or run swarm-rd-cli mcp and connect to it as an MCP server. Both give structured, parseable output.

Is this a library or just a CLI?
Both. event_log.py's EventLog and EventLogActor classes are directly importable if you're already in Python and don't need the CLI or MCP layer.

What license is this under, and can I use it commercially?
Apache 2.0. Commercial use, modification, and redistribution are all permitted under its terms; see LICENSE for the full text.

Contributing

There's no CONTRIBUTING.md yet since this is a pre-validation spike, not an accepted-contributions project. Open an issue if you want to discuss a change before this exists formally.

License

Apache 2.0. See LICENSE for the full text.

常见问题

6 问
答案由本站依据条目字段整理,不引用上游原文
Swarm Rd Orchestrator 怎么安装?

资料没有给出安装命令,请到官方仓库或文档按说明接入;这类条目多数是生态应用,或尚未发布可安装包。

Swarm Rd Orchestrator 怎么在 Claude Desktop、Claude、Cursor 里配置?

它的传输方式未标注,在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,配置才会生效。

Swarm Rd Orchestrator 收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于服务本身和背后的接口额度,正式使用前按官方说明确认。

Swarm Rd Orchestrator 需要 API Key 吗?

资料没有标注鉴权方式,是否需要密钥取决于服务本身,接入前按官方文档确认。

Swarm Rd Orchestrator 支持哪些客户端?

资料标注支持 Claude Desktop、Claude、Cursor、VS Code、Gemini CLI、Codex;其他客户端只要能配置 MCP Server,也能用同样的命令或地址接入。

Swarm Rd Orchestrator 能提供哪些能力?

资料显示能力标签包括 记忆。具体工具清单以官方文档为准。

热门专题

按标签挑服务器
全部专题 →

服务器信息以作者仓库与官方文档为准;分类、传输方式与鉴权方式的中文说明为中文整理,最近核验 未标注。