tomeraitz/claude-slack-bridge

开发工具

本页由 AI 翻译，技术细节请以原文为准。查看原文

35 stars 16 forks 质量 75 趋势 26MIT

Repository Install

MCP服务器允许Claude Code在任务中暂停，并通过Slack向真人提问——回复后再继续。

概览

Claude Code 与 Slack 之间的双向桥梁： - Claude 在任务中途暂停，通过 Slack 提问，等待你的回复后继续。- 在 Slack 频道中标记机器人，Claude 会根据频道上下文完整运行——它知道该做哪个项目。 - — Claude 在任务中途暂停，向 Slack 发布问题，阻塞直到你在线程中回复，然后继续。多个并发会话由 thread_ts 正确路由。- — Slack 频道中的 @claude-bot 在容器内匹配的项目目录中生成 claude -p。通过 [label] 前缀支持 git 工作树。- — 运行没有墙时限限制;它工作直到完成或你停止运行。

概述（AI 翻译）

Claude Code 与 Slack 之间的双向桥梁：

Claude → Slack： Claude 在任务中途暂停，通过 Slack 提问，等待你的回复，然后继续。
Slack → Claude： 在 Slack 频道中标记机器人，Claude 会根据频道知道该做哪个项目。

它的作用

ask_on_slack MCP 工具 — Claude 在任务中暂停，向 Slack 发布问题，阻塞直到你在线程中回复，然后继续。多个并发会话由 thread_ts 正确路由。
项目感知Slack机器人 — Slack通道中的@claude-bot在容器内匹配的项目目录中生成claude -p。通过[label]前缀支持git工作树。
实时进度，无超时 — 跑动没有墙壁计时限制;它一直有效，直到完成或你停止。

本页由 AI 翻译，技术细节请以原文为准。

README

以下为原文 README 节选，完整内容请查看原文。

Claude ↔ Slack Bridge

A two-way bridge between Claude Code and Slack:

Claude → Slack: Claude pauses mid-task, asks a question via Slack, waits for your reply, and resumes.
Slack → Claude: Tag the bot in a Slack channel and Claude runs with full project context — it knows which project to work on based on the channel.

Claude Code  ──ask_on_slack──▶  Slack channel  ──your reply──▶  Claude Code resumes
Slack @bot   ──────────────────▶  claude -p (in project dir) ──▶  reply in thread

What It Does

ask_on_slack MCP tool — Claude pauses mid-task, posts a question to Slack, blocks until you reply in the thread, then resumes. Multiple concurrent sessions are routed correctly by thread_ts.
Project-aware Slack bot — @claude-bot in a Slack channel spawns claude -p in the matching project directory inside the container. Supports git worktrees via a [label] prefix.
Live progress, no timeout — a run has no wall-clock limit; it works until it finishes or you stop it. While it runs, the bot edits one status message in the thread: a todo checklist, recent actions (newest first, up to 99/update), a changed-files section with per-file +/− churn, and a grey resource line (per-tool breakdown, subagents, skills used, MCP servers, compaction count, errors, tokens in/out/cache + estimated cost, context used/window + %, turns, model, elapsed) — all derived from the CLI’s event stream with no extra model calls. Actions are numbered (newest = highest, on top). A quiet run shows an ⏳ idle hint (IDLE_HINT_SECONDS); a genuinely stuck one (no activity for IDLE_TIMEOUT_SECONDS) is killed. When it finishes, the message keeps the full panel with a ✅ Done headline on top (the whole greyed section stays — nothing is collapsed away). Per-model context windows and prices live in model_context.json. Turn the live message off with LIVE_PROGRESS=false (stop then falls back to the 🛑 reaction). Preview the layout locally with tools/preview_status.py.
Stop a running task — react with 🛑 to the live status message (the bot pre-adds the 🛑 there) or to the message that started the run. Either way the bot kills it (and everything it spawned), posts ⏹️ Stopped., and drops the partial reply. No Slack Interactivity setup needed — it’s a plain reaction.
Full-process plugin — a turnkey feature-development workflow driven from Slack (/process start → pick a task → worktree → design → plan → run-plan → PR per step).

Quickstart — Claude → Slack

1. Create a Slack app and get tokens

Follow docs/slack-setup.md to create a Slack app, get your xoxb- and xapp- tokens, and invite the bot to a channel.

在 GitHub 查看完整 README

安装

docker exec -i -e SLACK_CHANNEL claude-slack-bridge python session.py

新手？先看 mcpskill.org 教程 →

配置

{
  "mcpServers": {
    "claude-slack-bridge": {
      "command": "docker",
      "args": [
        "exec", "-i",
        "-e", "SLACK_CHANNEL",
        "claude-slack-bridge",
        "python", "session.py"
      ],
      "env": {
        "SLACK_CHANNEL": "#your-project-channel"
      }
    }
  }
}