ChatAtlas Local

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

将你的 ChatGPT 和 Claude 数据导出导入到一个本地 SQLite 文件中,并为 AI 助手提供只读工具来搜索和读取这些数据。无需网络。

一句话选型结论

按资料字段整理

ChatAtlas Local 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
ChatAtlas Local当前 未标注 未标注 未标注 无 2026-11-20
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2026-11-20上游最后核验时间

详细介绍

11.5 KB
来自项目仓库与上游目录的原文,未做删改

ChatAtlas Local

Your ChatGPT and Claude chats, searchable on your own computer.

Ask Claude about them. Nothing leaves your machine.

You've had hundreds of conversations with AI assistants. The good answers, the decisions, the
half-finished plans are scattered across ChatGPT and Claude, and their own search barely helps.

ChatAtlas Local takes the data exports both of them already give you and keeps every chat in one
file on your computer. Then you can:

  • Search everything at once, from the terminal or in your browser.
  • Ask Claude about it: "what did ChatGPT tell me about my sourdough starter?" or "find the chat
    where I planned the garden beds." Claude Desktop, Claude Code, Cursor, VS Code and others can
    read your archive through MCP.
  • Keep it forever, as plain Markdown files or a backup, whatever happens to the apps.

No account, no server, no telemetry. It never connects to the internet.

Get started

You need Node.js 22.16 or newer (or 24). Then:```sh
npm install -g chatatlas-local


Before the first npm release, install it straight from GitHub: `npm install -g github:hnt504/chatatlas-local`

### 1. Ask for your exports

- **ChatGPT:** Settings → Data controls → Export data. You get an email with a link to a .zip.
- **Claude:** Settings → Privacy → Export data. Same idea: an email, then a .zip.

### 2. Bring them in```sh
chatatlas-local import ~/Downloads/<your ChatGPT export>.zip ~/Downloads/<your Claude export>.zip
``````
✓ 6720ab…-2026-10-21.zip: ChatGPT, 1,284 chats (1,284 new)
✓ data-2026-10-21.zip: Claude, 412 chats (412 new)

Do it again whenever you like with a newer export: new chats come in, changed ones update, nothing
doubles up. Or let it happen by itself (see Exports folder).

One Claude chat, every branch. On claude.ai, Retry or editing a message hides the old reply
behind a small version switcher, and the next sync of most tools keeps only what is on screen. To
save a chat with all of its branches, open it in Chrome, press F12, choose Console, and paste:``js (async () => { const id = location.pathname.match(/\/chat\/([0-9a-f-]+)/)[1]; const [org] = await fetch('/api/organizations').then((r) => r.json()); const chat = await fetch(/api/organizations/${org.uuid}/chat_conversations/${id}?tree=True&rendering_mode=messages&render_all_tools=true).then((r) => r.json()); const a = Object.assign(document.createElement('a'), { href: URL.createObjectURL(new Blob([JSON.stringify(chat)])), download:claude-${id}.json` });
a.click();
})();


Then `chatatlas-local import claude-.json`. The branch on screen comes first; each hidden
branch follows it, opening with "Earlier version, hidden on claude.ai by a retry or an edit."

### 3. Look around```sh
chatatlas-local search sourdough starter
chatatlas-local open            # the browser view in the screenshot above

Connect Claude (and friends)

Claude Desktop

Either download chatatlas-local-.mcpb from the
latest release and open it with
Claude Desktop (it asks where your exports go; no terminal needed), or, with the command
installed:```sh
chatatlas-local setup claude-desktop --write


Quit Claude Desktop completely and open it again. Then ask it something like *"Search my old chats
for the kitchen remodel estimate."*

**About the red warning.** When you open the extension, Claude Desktop says installing it "will
grant this extension access to everything on your computer." It says that for every extension
installed from a file rather than from Anthropic's directory, because an extension is a program
that runs as you. Here's what this one actually does: it reads your exports and its own archive
file, never connects to the internet, and gives Claude read-only tools. The code is all here to
check.

### Claude Code

Install it as a plugin (the MCP tools and a skill that teaches Claude when to look here):```
/plugin install chatatlas-local --marketplace hnt504/chatatlas-local

Or, with the command installed:```sh
chatatlas-local setup claude-code --write # the MCP tools, for every project
chatatlas-local skill # teaches Claude Code when to look here


### Cursor, VS Code, Windsurf, Gemini CLI, Codex```sh
chatatlas-local setup                 # shows each app's settings, and whether it's connected
chatatlas-local setup cursor --write  # or vscode, windsurf, gemini-cli, codex

--write keeps a copy of the app's old settings file next to it before changing anything.

Assistants in a browser page (llama.cpp, Open WebUI)

Some assistants run inside a web page and can't start a program the way Claude Desktop does. They
connect to an MCP address instead. mcp --http gives them one, on your computer only:```sh
chatatlas-local mcp --http --allow-origin http://localhost:8081


Then add `http://127.0.0.1:8770/mcp` as an MCP server in the assistant's settings (in llama.cpp's
web UI, that's the MCP Servers page). Use the address its page opens at for `--allow-origin`. With
a local model, your chats never leave your computer.

It listens on `127.0.0.1` only, answers only requests addressed to it, and lets in only the pages
you name with `--allow-origin` (several are fine, separated by commas). Programs on your computer
that aren't browsers can use it too, as they could read the archive file anyway. Use `--port` for
another port, and `--verbose` to see each request (never your chats or search words). Stop it with
Ctrl+C.

Here the tools are named `chatatlas_search_chats`, `chatatlas_read_chat` and so on. Browser
assistants put every tool in one list, and some have a `search_chats` of their own (llama.cpp's
web UI does, for chats saved in the web UI), which would otherwise be used instead.

### What about ChatGPT?

ChatGPT can't reach a file on your computer, so it can't read your archive here. You can still
import your ChatGPT exports and search them from Claude, the terminal or the browser view.

## What the assistant can do

Everything is read-only. An assistant can find and read your chats; it can't change, tag or delete
anything.

| Tool | What it does |
| --- | --- |
| `search_chats` | Chats holding every word you asked about, with the best-matching message quoted |
| `list_chats` | Browse by platform, date, words in the title, or tag |
| `read_chat` | Read a conversation message by message (long ones a page at a time) |
| `recent_topics` | What you talked about over a stretch of days, week by week |
| `archive_overview` | What the archive holds and how far back it goes |

## An exports folder

Point ChatAtlas Local at a folder (your Downloads, say) and any ChatGPT or Claude export you save
there comes in by itself. Other files are left alone, and each file is only read once.```sh
chatatlas-local watch ~/Downloads                              # until you press Ctrl+C
chatatlas-local setup claude-desktop --inbox ~/Downloads --write  # whenever Claude Desktop is open

The Claude Desktop extension asks for this folder when you install it.

Commands

Command
import A ChatGPT or Claude export (the .zip, its conversations.json, or the unzipped folder), or a ChatAtlas backup
search Chats holding every word. Each word also finds longer words it starts ("plan" finds "planning"); accents don't matter
list Newest first
show Read one chat. --md for Markdown, --json for the data. The start of an id is enough
topics What you talked about lately (--days 90)
stats Chats per platform, chats per month, your tags
open The browser view. --read-only, --port 8765, --no-browser
tag Tag a chat (--remove takes tags off)
forget Take chats out of the archive
export Every chat as a Markdown file, or --format chatatlas for one backup file
watch Bring in exports saved there, and keep watching
setup [app] [--write] Connect an assistant
skill Install the Claude Code skill
doctor Check Node, the archive, the search index and every connected app
where Where the archive file is

Filters for search, list, export and topics: --platform chatgpt, --tag work,
--since 2025-03, --until 2026, --title garden, --limit 50, --offset 50, --json.

Privacy policy

ChatAtlas Local is a program that runs on your computer. It has no servers and no accounts.

  • What it collects: nothing. It has no analytics, telemetry or crash reporting, and never makes
    a network connection.
  • What it stores, and where: the chats from the exports you import, plus any tags you add, in
    one SQLite file on your computer: ~/.chatatlas-local/chatatlas.db (chatatlas-local where
    says exactly where). Use another with --db or the CHATATLAS_LOCAL_DB setting. It reads
    your export files and doesn't change or delete them.
  • Sharing: ChatAtlas Local shares nothing with anyone, including its author. If you connect an
    AI assistant (Claude Desktop, Claude Code, Cursor and so on), that assistant can search and read
    your archive through read-only tools when you ask it to, and what it reads becomes part of your
    conversation with it, under that assistant's own privacy policy.
  • The browser view listens on 127.0.0.1 only, answers only requests addressed to it, and every
    request needs a random key that only the tab it opened knows.
  • mcp --http listens on 127.0.0.1 only, answers only requests addressed to it, and refuses
    any web page you didn't name with --allow-origin before it reads anything.
  • How long it's kept: until you delete it. chatatlas-local forget removes a chat;
    removing the ~/.chatatlas-local folder removes everything. Your ChatGPT and Claude accounts are
    never touched.
  • Installing: npm (or GitHub, for the Claude Desktop extension) serves the download under its
    own terms; after that, nothing goes over the network.
  • Contact: questions go to GitHub issues;
    anything that could expose someone's chats goes privately through
    security advisories.

ChatAtlas Local and ChatAtlas Pro

ChatAtlas Local is free and open source, and stays that way.
ChatAtlas Pro is the hosted version, for people who want more:

Local Pro
ChatGPT and Claude exports ✓ ✓
Search by words ✓ ✓
Claude Desktop, Claude Code, Cursor and more ✓ ✓
Your data stays on your computer ✓ Kept in your account
Gemini, Grok, Perplexity, DeepSeek and more Through a ChatAtlas backup ✓
Search by meaning ✓
ChatGPT and claude.ai on the web ✓
Folders, projects, any device ✓

They speak the same backup format: chatatlas-local export --format chatatlas makes a
file ChatAtlas Pro imports, and a ChatAtlas backup comes into ChatAtlas Local the same way.

Questions

Which Node do I need? 22.16 or newer, or 24 and up (23 won't do), because ChatAtlas Local uses the SQLite database built
into Node. chatatlas-local doctor checks.

A chat I had yesterday isn't there. The archive holds what was in the exports you imported.
Ask for a fresh export and import it (or drop it in your exports folder).

Can it read exports from Gemini or other apps? Not directly yet. ChatAtlas backups can carry
chats from any platform, and support for more exports is welcome as a pull request.

How big does it get? About the size of the exports' text. A 60 MB export of 3,000 chats
imports in about two seconds and searches in well under a tenth of a second.

Contributing

Bug reports and pull requests are welcome. See CONTRIBUTING.md.

License

MIT. Made by Heather Turner, who also makes ChatAtlas.

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
ChatAtlas Local 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。

ChatAtlas Local 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。

ChatAtlas Local 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。

ChatAtlas Local 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。

ChatAtlas Local 支持哪些客户端?

资料标注支持 Claude Desktop、Claude、Cursor、VS Code、Windsurf、Gemini CLI;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。

ChatAtlas Local 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 2026-11-20。