Venom
io.github.venomlabs-dev由 GitHub 帳號 venomlabs-dev 發布v1.0.0更新於 Nov 14, 2026
Multiplayer for the coding agents your team already uses: shared tasks, decisions, skills and tools.
概覽
遠端 MCP 伺服器,讓團隊的編碼代理在 Claude Code、Codex、Cursor 等用戶端之間共用專案、任務、決策與技能。
- 可以做什麼
- 讀取並更新團隊的 Venom 工作區:專案、任務與決策
- 共用技能與工具,讓其他代理或同事接手完成工作
- 透過伺服器回傳的指引讓目前任務保持最新
- 在多個 MCP 主機中以同一個組織範圍的登入連線同一個工作區
- 適用情境
- 適合多人或多個代理協作同一份程式碼庫、希望交接內容、決策與任務狀態能跨工作階段、跨不同編碼代理保留,又不想更換現有工具鏈的情境。
- 執行需求
- 遠端端點 mcp.venomlabs.dev,不需要本機套件或執行環境。需要 Venom 帳號,並在每個主機完成 OAuth 授權,連線限定於所屬組織。各主機設定不同:Claude 與 Codex 以外掛安裝,Gemini CLI 使用擴充目錄,Cursor、Kiro、OpenCode、Copilot、Cline、Windsurf、Devin 需合併 MCP 設定並加入說明檔。30 天未使用需重新登入。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Venom,將其新增到工作區。
- 連接 所需的 憑證。它們會被私密儲存,絕不會在公開頁面上展示。
- 為需要使用其工具的對話啟用該服務。
Web executable,透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。
交給代理程式安裝
把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式,或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南,先向你展示發布者、端點和登入方式,經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。
閱讀 https://sourceweft.com/mcp/io.github.venomlabs-dev%2Fvenom/install.md,並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前,先告訴我它的發布者、端點和登入方式,等我確認。不要把 API 金鑰寫進命令或我會提交的檔案;問我金鑰應從哪裡取得。其他 MCP 客戶端
把它新增到你客戶端的 mcpServers 設定中。 客戶端會在首次連接時完成 OAuth 登入。
{
"mcpServers": {
"venom": {
"type": "http",
"url": "https://mcp.venomlabs.dev/"
}
}
}README
Venom MCP plugin
Venom AI by Venom Labs makes the coding agents your team already uses multiplayer. One MCP server connects Claude Code, Codex, Cursor, Gemini CLI and other MCP clients, so your team's agents share tasks, decisions, skills and tools. There is no new harness to adopt.
Website · How it works · Quickstart · MCP setup · Pricing
Venom is where your team's AI agents share their work: projects, tasks,
decisions and skills, so any agent or teammate can pick up where another left
off. Agents reach it through the host-managed MCP connection at
https://mcp.venomlabs.dev.
The plugin contains connection settings, a venom skill, and host adapters.
In Claude and Codex, the skill and the Venom server's own instructions and tool
responses guide the agent. Other hosts use the short VENOM.md guidance as a rules
file. Connections remain
authenticated and organization-scoped; the host manages OAuth and tool
approval. Installing this plugin does not disable those controls.
What this plugin does
Venom connects Claude to your team's Venom workspace through one remote MCP
server, https://mcp.venomlabs.dev, which you sign in to with your Venom
account. The plugin ships a venom skill that tells Claude when to read and
update Venom projects and tasks. It registers no hooks and runs no
code on your machine. Claude sends data to Venom only through the Venom tools it calls, and
only to your organization's workspace.
Claude (chat, Cowork and Claude Code)
Add Venom from the plugin directory on claude.ai, or in Claude Code:
Authorize Venom when prompted. In chat and Cowork, connect Venom on the
plugin's Connectors tab. The venom skill loads when a request concerns
work, the server's instructions reach Claude Code and Cowork at session start,
and each Venom tool response reminds Claude to keep the task current.
Codex
Start a new task and authorize Venom. The plugin ships the same venom skill
as Claude and registers no hooks, because OpenAI's plugin directory does not
accept plugins with lifecycle hooks. The server's instructions and a reminder
on each Venom tool response keep the task current.
Venom Full (for developers)
venom-full adds hooks to venom. Anthropic's and OpenAI's plugin directories
don't accept hooks from us, so it isn't listed there; install it from this
marketplace instead. It needs venom, ships only hooks, and uses the venom
plugin's connection, so there's one sign-in and one set of Venom tools. Its
files live in venomlabs-dev/venom-full;
the marketplace fetches them for you.
Claude Code installs venom with it:
Codex needs both plugins and lifecycle hooks turned on:
Then trust the Venom hooks once in /hooks.
The hooks are mcp_tool hooks that call the Venom server's get_guidance
tool after compaction, before compaction (Claude Code), when a subagent starts,
on each prompt, and after git commit, git push, gh pr create and
gh pr merge. No local program runs. While the hooks are active, the server
stops adding reminder lines to Venom tool responses.
Gemini CLI
Install the extension folder from a clone of this repository:
Restart Gemini and authorize Venom. The extension loads its VENOM.md as a
context file in every session. To update, pull this repository and run
gemini extensions update venom. Do not install the repository root as a
Gemini extension.
Other hosts
Merge MCP entries into existing configuration; never overwrite other servers. Add the small instruction file alongside existing rules.
Authorize each connection in its host. Use current supported host releases. The coverage notes describe sources, lifecycle limits, and verification. These are local/IDE adapters; unattended cloud agents need a separately supported authentication path.
Troubleshooting
If your agent says Venom isn't connected, or never uses Venom, sign in again:
run /mcp and authenticate venom in Claude Code, run codex mcp login venom
for Codex, or reconnect Venom on the plugin's Connectors tab in Claude chat
and Cowork. A sign-in lasts while you use it; after 30 days without use you
sign in again.
Upgrading to 0.8.1
The Codex manifest now carries the review test cases and release notes for OpenAI's plugin directory. Nothing changes for users.
Upgrading to 0.8.0
The marketplace now also lists venom-full, hooks for developers: see
Venom Full. The venom plugin is unchanged.
Upgrading to 0.7.1
The venom skill now tells the agent what to do when Venom isn't connected:
say so in one line, with the sign-in step for your app, instead of working
without Venom unnoticed.
Upgrading to 0.7.0
The Codex plugin no longer ships hooks, so it can be listed in OpenAI's plugin
directory. Codex now loads the venom skill and gets reminders on Venom tool
responses, as Claude does. You can remove the old Venom hooks from /hooks.
The venom skill now loads before the first reply in any conversation about
your team's work, including research and questions, not only when building.
Without hooks, the skill is what starts Venom in chat and Cowork, where the
model doesn't see the server's instructions.
Upgrading to 0.6.0
The Claude plugin no longer ships hooks. Anthropic's plugin directory accepts
hooks only from Verified organizations, so Claude gets Venom guidance from the
venom skill, the server's instructions, and Venom tool responses instead.
Codex keeps its hooks, now in .codex-plugin/hooks.json; trust them again via
/hooks. Gemini installs from platforms/gemini and no longer runs a
SessionStart hook. Refresh manually copied rules files from this release.
Development
The repository holds only Markdown and JSON, as the directory requires for
organizations that aren't Verified. Edit VENOM.md, then update the
copies under platforms/ to match. The checks for this repository live in the
Venom monorepo.
來源:README.md,提交 8c7a3b0
工具
0版本歷史
1- v1.0.0最新Nov 14, 2026
評分
已滿足適用檢查項的 85%
該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。
- 近 30 天內可連通必需25 分
- 已宣告認證方式必需10 分
- 已列出工具15 分
請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。
- 描述必需15 分
- 圖示10 分
- 文件15 分
- 授權條款10 分