Zulip MCP D244adfc

MCPAI代理

提供方 adamhajari

AI代理 stdio · 本地进程 无需鉴权 社区 / 第三方 Claude DesktopClaude
查看接入方式

一个MCP服务器,允许Claude访问您的Zulip频道,以便获取最近活动的摘要。

一句话选型结论

按资料字段整理

Zulip MCP D244adfc 属于AI代理类 MCP 服务器,stdio · 本地进程,无需鉴权,可直接接入 Claude Desktop、Claude,有可直接执行的安装命令。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AI代理 传输方式stdio · 本地进程 鉴权无需鉴权 部署方式本地部署
GitHub 星标 0 仓库热度,无公开仓库时显示未标注
工具数 6 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

AI代理
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6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

聊天摘要Python本地部署自动化工具隐私保护本地LLM集成

安装与配置

stdio · 本地进程 无需鉴权
客户端在本机拉起子进程,数据不出本机
python prompt_with_privacy.py "your query here"

这条命令只是入口,具体启动参数请按官方文档补齐,不要直接照搬到别的服务器。

  1. 1
    准备运行环境

    stdio 服务在本地起进程,需要先装好 Python,再按安装命令拉取依赖。

  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

    连接成功后客户端会列出约 6 个工具,随便调一个只读工具确认能返回结果。

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
给代理接外部能力

让代理在任务中途调用搜索、数据库等真实工具。

多步任务编排

把一串操作拆成步骤,由代理按顺序执行。

自动化重复流程

把固定流程交给代理跑,减少手工操作。

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

  • 资料标注 6 个工具,职责相对集中,接入后不容易误操作。
  • 有现成的安装命令,不用自己找入口,照着跑就能起服务。
  • 已适配 Claude Desktop、Claude,换客户端不用重新研究配置。
  • 资料标注无需鉴权,装好就能连。

先想清楚的地方

  • stdio 方式要求本地装好运行时(如 Node.js),换机器要重新装一遍。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在AI代理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

  • 完全不想在本地装环境的人。
  • 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
Zulip MCP D244adfc当前 6 stdio · 本地进程 本地部署 有 2026-05-17
Filesystem 560c4363 13 未标注 本地部署 无 2026-05-17
Server Everything (everything) 11 stdio · 本地进程 本地部署 有 2026-05-17
Time 7365cb6d 2 SSE · 服务端推送 远程部署 有 2026-05-18
Everything Claude Code 未标注 stdio · 本地进程 本地部署 有 2026-05-17
Openmemory 557e1585 未标注 stdio · 本地进程 本地部署 有 2026-05-17

服务器档案

AI代理
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adamhajari上游标注的维护方 adamhajari
分类AI代理本站按上游分类展示
连接方式stdio · 本地进程客户端在本机拉起子进程,数据不出本机
鉴权方式无需鉴权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 · Python本地安装时用到的包标识与运行时
部署类型本地部署上游给出的部署形态
最近核验2026-05-17上游最后核验时间

详细介绍

8.5 KB
来自项目仓库与上游目录的原文,未做删改

zulip-mcp

An MCP server that gives Claude access to your Zulip channels, so you can ask for summaries of recent activity.

Setup

1. Get your Zulip API key

  1. In Zulip, go to Settings → Account & privacy → API key
  2. Click Generate API key and copy it

2. Create a .zuliprc file

Copy the example file and fill in your credentials:```bash
cp .zuliprc.example .zuliprc


Then edit `.zuliprc` with your email and API key:```ini
[api]
key=your_api_key_here
email=your_email@recurse.com
site=https://recurse.zulipchat.com

Alternatively, Zulip can generate a pre-filled .zuliprc for you: Settings → Account & privacy → API key → Show/change your API key → Download .zuliprc. Place the downloaded file in the project directory.

3. Configure your channels

Edit config.yaml to set the channels you want in your default digest and how far back to look:```yaml
zuliprc: .zuliprc

defaults:
hours_back: 24
channels:
- general
- announcements
- your-interests


Use exact stream names as they appear in Zulip.

## 🛡️ Architecture & Data Privacy
This is a standard MCP server. It is LLM-agnostic and can be used with any MCP-compatible client. The privacy of your Zulip data depends entirely on the client you choose:
  - Cloud Clients (e.g., Claude Desktop): Data will be sent to the provider's servers for processing.
  - Local Clients (e.g., Ollama + local MCP host): Data stays on your machine.
If you are using this in a community with high privacy expectations, consider using a local-only workflow.

### 4. Claude Install

If you're using Claude Code, you can have Claude handle steps 4 and 5 for you. Open Claude Code in the project directory and paste this prompt:

> "Install dependencies for this project and register it as an MCP server in my Claude Code settings."

### 5. Install dependencies (manual)```bash
uv sync

6. Add to Claude Code (or Claude Desktop) (manual)

Claude Code — run from the project directory:```bash
claude mcp add zulip-mcp -- uv run --project /absolute/path/to/zulip zulip-mcp


This writes the server into `~/.claude/settings.json`. To scope it to this project only, add it manually to `.claude/settings.json` (committed) or `.claude/settings.local.json` (local only) in the project root:```json
{
  "mcpServers": {
    "zulip-mcp": {
      "command": "uv",
      "args": ["run", "--project", "/absolute/path/to/zulip", "zulip-mcp"]
    }
  }
}

Claude Desktop — add to ~/Library/Application Support/Claude/claude_desktop_config.json:```json
{
"mcpServers": {
"zulip-mcp": {
"command": "uv",
"args": ["run", "--project", "/absolute/path/to/zulip", "zulip-mcp"]
}
}
}


> **Note:** The server reads `config.yaml` from the project directory by default. You can override this with the `ZULIP_MCP_CONFIG` environment variable:
>```json
> {
>   "mcpServers": {
>     "zulip-mcp": {
>       "command": "uv",
>       "args": ["run", "--project", "/absolute/path/to/zulip", "zulip-mcp"],
>       "env": {
>         "ZULIP_MCP_CONFIG": "/absolute/path/to/config.yaml"
>       }
>     }
>   }
> }
> ```

## Usage

Once the MCP is connected, you can ask things like:

- *"Summarize what's been discussed in #general today"*
- *"Give me a digest of all my configured channels"*
- *"What topics came up in #announcements in the last 48 hours?"*
- *"List my subscribed channels"*

To protect user privacy, pass `anonymize=true` to any tool that returns message content. Real names will be replaced with stable aliases (`{User1}`, `{User2}`, etc.) before anything is sent to Claude:

- *"Summarize the checkins channel from the last 24 hours. Use anonymize=true for all tool calls."*

The alias mapping is saved locally at `~/.config/zulip_mcp/anonymizer_map.json` for de-anonymization after the fact.

### Privacy-first CLI: `prompt_with_privacy.py`

For a workflow that anonymizes messages before sending to the LLM and deanonymizes the response, use the included script. It sends your query to the LLM with anonymization enforced, then automatically de-anonymizes the response using the local alias map before writing the result to a file.```bash
python prompt_with_privacy.py "your query here"

⚠️ Privacy & Security Disclosure

Please be aware of the following before using it in a community setting (like the Recurse Center):

  • Data Processing: By default, the prompt_with_privacy.py script sends Zulip message content to Anthropic's servers for processing. Even when using the anonymize=true flag, users may still be identifiable through writing patterns, specific interests, or context (stylometry).
  • Consent: Consider getting the consent of all participants before using this tool to summarize any non-public Zulip channels.
  • Local Alternative: If you require 100% data privacy, it is recommended to modify the client to point to a local LLM (e.g., via Ollama or an internal cluster like RC's Heap) rather than using a cloud-based provider.

Options:

Flag Default Description
--output, -o output.md Path to write the de-anonymized response

Example:```bash
python prompt_with_privacy.py "summarize checkins from the last 24 hours. return a one sentence summary for every person." --output checkins.md


The script:
1. Appends *"Use anonymize=true for all tool calls."* to your prompt
2. Invokes `claude -p` with the Zulip MCP and no other tools
3. Replaces all `{UserN}` aliases in the response with real names
4. Writes the result to the output file

> **Note:** Requires `claude` to be available in your `PATH` and the `zulip-mcp` server registered in `~/.claude.json` for this project directory (see setup steps above).

### Local LLM CLI: `zulip_local.py`

For a fully local workflow that keeps all Zulip data on your machine, use the included script. It connects directly to a local Ollama instance (or any OpenAI-compatible endpoint) and runs the full agentic loop — fetching messages and summarizing them — without sending anything to a cloud provider.

#### 1. Install Ollama

Download and install from ollama.com, or on macOS:```bash
brew install ollama

2. Download a model```bash

ollama pull gemma4:e4b


Any model that supports tool calling will work. `gemma4:e4b` is the default.

#### 3. Start the Ollama server```bash
ollama serve

Or run it as a background service via Homebrew (starts automatically on login):```bash
brew services start ollama


Either way, the server listens at `http://localhost:11434`.

#### 4. Run the script```bash
uv run python zulip_local.py "summarize the checkins channel from the last 24 hours"

Configuration

Environment variable Default Description
ZULIP_MCP_MODEL gemma4:e4b Model to use
ZULIP_MCP_DIR Script directory Path to the zulip-mcp project
OPEN_WEBUI_API_KEY — Set to use Open WebUI instead of local Ollama
OPEN_WEBUI_BASE_URL — Base URL for Open WebUI (required with OPEN_WEBUI_API_KEY)

Both OPEN_WEBUI_API_KEY and OPEN_WEBUI_BASE_URL must be set to switch away from local Ollama.

CLI flags

Flag Description
--model Override the model (takes precedence over ZULIP_MCP_MODEL)
--verbose Print tool calls and raw tool results

Example:bash uv run python zulip_local.py "summarize the checkins channel from the last 24 hours. give a one to two sentence summary for every person." --model llama3.2

Available tools

Tool Description
list_channels List all channels you're subscribed to
get_channel_messages Fetch recent messages from one channel
get_checkins Fetch check-ins one topic at a time (avoids context overload)
get_digest Fetch messages from all configured channels at once
get_full_message Fetch the complete content of a single message by ID
set_interests Save your interests to config.yaml to filter digests

All tools that return message content accept an optional anonymize parameter (default: false).

You can override defaults inline — e.g. "summarize #checkins for the last 3 days" will pass hours_back: 72 to the tool.

Notes

  • Messages longer than 500 characters are truncated in digest/channel views. Truncated messages include their ID and a note, so Claude can call get_full_message to retrieve the complete content when it seems relevant.
  • The server fetches up to 1000 messages per channel per request.
  • .zuliprc contains your API key — don't commit it. It's in .gitignore by default.

常见问题

8 问
答案由本站依据条目字段整理,不引用上游原文
Zulip MCP D244adfc 怎么安装?

在客户端按下面的安装命令接入:python prompt_with_privacy.py "your query here"。装完在客户端配置里引用同一个包或端点。

Zulip MCP D244adfc 怎么在 Claude Desktop、Claude 里配置?

它的传输方式是stdio · 本地进程,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。

Zulip MCP D244adfc 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。

Zulip MCP D244adfc 需要 API Key 吗?

资料标注为无需认证,装好就能连;如果实际调用时报鉴权错误,按官方文档补上凭据。

Zulip MCP D244adfc 支持哪些客户端?

资料标注支持 Claude Desktop、Claude;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。

Zulip MCP D244adfc 能提供哪些能力?

资料显示约 6 个工具;能力标签包括 Python、Claude、隐私保护。具体工具清单以官方文档为准。

Zulip MCP D244adfc 需要什么运行环境?

运行时为 Python;属于本地进程,需要在本机装好运行环境。

Zulip MCP D244adfc 是本地还是远程?

它是本地进程(stdio),由客户端在本机拉起。传输方式为stdio · 本地进程。

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 2026-05-17。