Moling

MCP浏览器工具

提供方 gojue

浏览器工具 未标注 OAuth 授权 社区 / 第三方 Claude DesktopClaudeCursorCline
查看接入方式

MoLing is a computer-use and browser-use based MCP server. It is a locally deployed, dependency-free office AI assistant.

一句话选型结论

按资料字段整理

Moling 属于浏览器工具类 MCP 服务器,传输方式未标注,OAuth 授权,可直接接入 Claude Desktop、Claude、Cursor 等,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。

分类浏览器工具 传输方式未标注 鉴权OAuth 授权 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

浏览器工具
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

agentaimcp

安装与配置

未标注 OAuth 授权
上游未给出传输方式
/usr/local/bin/moling

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

    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
自动操作网页

打开页面、点击、填表,把浏览器动作交给模型执行。

抓取结构化数据

把页面内容整理成表格或 JSON,省掉手工复制粘贴。

关键流程回归

按步骤检查页面是否正常,回归时少漏环节。

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

  • 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、Cursor,换客户端不用重新研究配置。

先想清楚的地方

  • 使用前要准备好OAuth 授权对应的凭据,并按最小权限配置。
  • 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

  • 需要在浏览器工具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

  • 不方便管理密钥或第三方授权的人。
  • 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
Moling当前 未标注 未标注 未标注 有 未标注
Playwright MCP Server 21 stdio · 本地进程 本地部署 有 2026-05-18
Fetch 9a2682de 1 SSE · 服务端推送 支持远程部署 有 2026-05-18
Browser 738d52bc 未标注 stdio · 本地进程 本地部署 有 2026-05-17
MCP 40600ece 未标注 stdio · 本地进程 本地部署 有 2026-05-17
Skyvern MCP 未标注 stdio · 本地进程 本地部署 有 2026-05-17

服务器档案

浏览器工具
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gojue上游标注的维护方 gojue
分类浏览器工具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OAuth 授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 · go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

18.5 KB
来自项目仓库与上游目录的原文,未做删改

About chromedp

Package chromedp drives browsers that speak the
Chrome DevTools Protocol from Go. It needs no external
driver.

Installing

Install the package with go get. The module needs Go 1.25 or newer. The JSON
types come from github.com/chromedp/cdproto/cdp/jsonv2. They are aliases of
the types of the standard library package encoding/json/v2 on Go 1.27 and
later, or on Go 1.25 and 1.26 with GOEXPERIMENT=jsonv2. Otherwise they are
aliases of the types of github.com/go-json-experiment/json. Version v0.20.0
uses the typed cdproto v0.157.8 and has the generic and iterator API. The
earlier versions of cdproto, v0.157.0, v0.157.1 and v0.157.2, have the old API.```sh
go get -u github.com/chromedp/chromedp


The module uses only the Go standard library and `cdproto`. It starts a
browser and talks to it through a pipe. Code that needs a websocket is in the
package `github.com/chromedp/chromedp/remote`, which has its own WebSocket
client. It is part of the same module, so `go get` installs it with the rest.
Import it only when you connect to a browser that runs already, when you want
the exec allocator to use a websocket, or when you keep the browser open after
the program ends.

## Usage

An action is a func that runs against a browser tab and returns a value.
`chromedp.Do` runs actions that return nothing, and `chromedp.Run` runs one
action and returns its value:```go
ctx, cancel := chromedp.NewContext(context.Background())
defer cancel()

if err := chromedp.Do(ctx, chromedp.Navigate(`https://pkg.go.dev/`)); err != nil {
    log.Fatal(err)
}
title, err := chromedp.Run(ctx, chromedp.Title())
if err != nil {
    log.Fatal(err)
}
fmt.Println(title)

A page event is an iterator. chromedp.Events subscribes when it returns, so
a program can subscribe, trigger the event, and then read it:``go loaded := chromedp.Events(ctx, page.LoadEventFired) if err := chromedp.Do(ctx, chromedp.Navigate(https://pkg.go.dev/`)); err != nil {
log.Fatal(err)
}
for ev, err := range loaded {
if err != nil {
log.Fatal(err)
}
fmt.Println(ev.Timestamp)
break
}


`docs/API.md` describes the API. It has 27 examples that show the
old code and the new code side by side. `docs/MIGRATION.md`
lists every renamed and removed name.

See the Go reference for the documentation and examples.

## More examples

The examples repository has 43 programs for larger tasks.
Each program is one `main.go` file. Run one with `go run`, for example
`go run github.com/chromedp/examples/tabs@latest`. Every program takes the flag `-v`
to print the protocol messages. Every program except `remote` takes the flag
`-visible` to show the browser window. Most programs take the flag
`-visible-on-terminal` to draw the page in the terminal while they run, with
`termcast`. Every program except `fast`
reads local content and needs no internet. Some programs take the flag `-url` to
read another site. The program `termcast` takes the flag `-fps` instead of
`-visible-on-terminal`.

Read pages and fill forms:

- click clicks an element that a selector finds.
- text reads the text of an element.
- eval runs JavaScript in the page and decodes the result.
- structeval decodes JavaScript results into Go structs, slices and maps, and shows the errors.
- logic mixes actions and Go code to read a list.
- subtree walks a subtree of the DOM.
- selectors shows the typed selectors side by side.
- frames reaches elements inside an iframe and a shadow root.
- submit fills out and submits a form.
- keys sends key events to an element.
- upload uploads a file on a form.
- visible waits until an element is visible.
- dragdrop drags and drops with the mouse and with HTML5 drag and drop.

Network, sessions and files:

- cookie sets cookies on requests.
- headers adds extra HTTP headers.
- proxy signs in to a proxy server that needs a password.
- intercept blocks, mocks and changes requests with the Fetch domain.
- session saves a login session and restores it in another browser.
- har writes a HAR file from the network events of a page.
- download_file downloads a file with a headless browser.
- download_image downloads an image from the network response.

Events and the protocol:

- eventsiter reads the events of a page with iterators, and waits for the network to be idle.
- console reads the console and the uncaught exceptions of a page.
- dialogs answers alert, confirm, prompt and beforeunload dialogs.
- popups works with popups and several targets, and with browser contexts.
- exposefunc calls Go functions from the page.
- extension loads a browser extension, uBlock Origin Lite, from a folder on disk and shows that it blocks the ads of a page.
- latlon drives a map and reads the latitude and the longitude from the URL of the page.
- rawcall sends protocol commands that have no action, such as timezone, locale, geolocation and throttling.

Screens and devices:

- screenshot takes a screenshot of an element and of the whole page.
- pdf prints a page to a PDF file.
- pdfoptions prints a page to PDF files with different options.
- pdfstream reads a printed PDF as a stream.
- emulate emulates a device, such as an iPhone.
- screencast saves the frames of a page as JPEG files.
- termcast plays an animated SVG and streams the screen of the browser to the terminal with `termcast`.

Several tabs and browsers:

- tabs uses several tabs of one browser, with or without a window for each tab, and switches between them.
- workers runs many jobs at the same time in one browser with a pool of goroutines.
- multi uses the headless-shell image in a container.
- remote connects to a browser that is already running, with the package `remote`.

The program fast reads a live site, `fast.com`, and draws an image in the terminal. It can fail when the site changes. The programs forecast and geoip also draw an image in the terminal. Run these programs in a terminal that can show images.
The tags of the examples repository follow the tags of `chromedp`.

## Show the browser in the terminal

`termcast` streams the screen of a `chromedp` page to a terminal
that can show images (Kitty, iTerm2 or Sixel). It uses the screencast of the
Chrome DevTools Protocol, so it works with a headless browser, also over `ssh`.
Start the browser first, and then start the stream with the context.```go
ctx, cancel := chromedp.NewContext(context.Background())
defer cancel()
chromedp.Do(ctx, chromedp.Navigate("https://example.com")) // starts the browser
s, err := termcast.Start(ctx)                               // draws 4 frames each second
if err != nil {
    log.Fatal(err) // termcast.ErrNoGraphics when the terminal has no graphics
}
defer s.Stop()

To draw one element and not the whole page, add
termcast.WithElement(chromedp.CSS("#chart")) to the call of termcast.Start.
The module termcast needs Go 1.25 or newer. The stream clears the terminal at
each redraw. It holds the log lines while it
runs, and prints them after the final frame. The type termcast.Flags adds the
flags -visible-on-terminal and -terminal-fps to a program, as the programs
in the examples repository have them.

Visible browser

By default, chromedp runs Chrome in headless mode, so no window opens. To see
the browser, add WithVisibleWindow. Set the variable CHROMEDP_VISIBLEWINDOW=1
to get the same result with no change in the code. On Linux, a visible window
needs DISPLAY or WAYLAND_DISPLAY. Without them, the first Run returns
chromedp.ErrNoDisplay.```go
ctx, cancel := chromedp.NewContext(context.Background(), chromedp.WithVisibleWindow())
defer cancel()
chromedp.Do(ctx, chromedp.Navigate("https://example.com"))
chromedp.WaitClosed(ctx) // block until the user closes the window


To leave the browser open after the program ends, add `remote.WithKeepOpen` from
the package `github.com/chromedp/chromedp/remote`. The program can print the
address, and another program can attach to it with `remote.NewAllocator`. The
profile directory stays on disk, and you must delete it yourself.```go
ctx, _ := chromedp.NewContext(context.Background(), chromedp.WithVisibleWindow(), remote.WithKeepOpen())
chromedp.Do(ctx, chromedp.Navigate("https://example.com"))
wsURL, profile := chromedp.KeptOpen(ctx)
fmt.Println("attach to", wsURL, "profile", profile)
// The program ends here. The browser stays open.

Remote browser

To connect to a browser that runs already, such as a container or a hosted
service, use remote.NewAllocator with the websocket address of the browser:```go
allocCtx, cancel := remote.NewAllocator(context.Background(), "ws://127.0.0.1:9222/")
defer cancel()
ctx, cancel := chromedp.NewContext(allocCtx)
defer cancel()
title, err := chromedp.Run(ctx, chromedp.Title())


A hosted service can need a header on the websocket request. Give the full
address of the browser and the options `remote.NoModifyURL` and
`remote.WithDialHTTPHeader`.

## Frequently Asked Questions

> I cannot see any Chrome browser window

By default, `chromedp` runs Chrome in headless mode. See
Visible browser. See also
`DefaultExecAllocatorOptions`, and see an example
that overrides the default options.

> I see "context canceled" errors

When the connection to the browser is lost, `chromedp` cancels the context. This
can cause the error. It happens, for example, when someone closes the browser by
hand, or when something kills the browser process. When the browser process
dies on its own, the error also holds the exit error of the process. Use
`errors.As` with an `*exec.ExitError` to read the signal or the exit status.

> How does chromedp talk to the browser it starts?

By default, through a pipe. `chromedp` starts Chrome with `--remote-debugging-pipe`
and two pipes, so Chrome opens no debugging port. On Windows the pipes are
handles that the switch `--remote-debugging-io-pipes` names. To use a
websocket and a debugging port instead, add the `remote.WebSocket` option of the
package `github.com/chromedp/chromedp/remote` to the exec allocator. A
`remote-debugging-port` or `remote-debugging-address` flag also selects the
websocket, and so does `KeepOpen`, and they all need `remote.WebSocket`. A
program for Windows needs no `remote` package for the default pipe.

> Chrome exits as soon as my Go program finishes

On Linux, `chromedp` kills the Chrome child processes that it started, so that no
resources leak. To leave Chrome open, add `remote.WithKeepOpen`. See
Visible browser. You can also start Chrome yourself and
connect with `remote.NewAllocator`.

> Calling an action or a command results in "invalid context"

`chromedp.Do`, `chromedp.Run` and `chromedp.Call` need a context that came from
`chromedp.NewContext`, because the context holds the browser and the tab. Any
other context gives `ErrInvalidContext`.

> How do I send a protocol command that has no action?

Call it with `chromedp.Call`, which runs the command on the tab of the context:```go
ctx, cancel := chromedp.NewContext(context.Background())
defer cancel()

res, err := chromedp.Call(ctx, page.GetFrameTree, cdp.Empty{})

Inside an action, call cdp.Call(ctx, t, command, params) with the target t
that the action receives. The target is the tab of the context. The commands and
their parameter structs are in github.com/chromedp/cdproto.

> I have an action of the old kind

Wrap it with chromedp.Legacy. The old kind is a value with a Do(context.Context) error
method.

> Is it safe to use one context from several goroutines?

Contexts that share one browser run in separate tabs, and they are safe to use
in parallel. Call chromedp.Run once on a parent context so that it has a
browser, and then make a child context for each goroutine with
chromedp.NewContext. A child of a context that has no browser yet starts a
browser of its own.

Do not share one context between goroutines. The actions of one context run in
one tab, so they can race. For example, two Navigate actions on the same tab
can fail. The first Run on a context starts the browser, and it must
not run at the same time as another Run on that context.```go
ctx, cancel := chromedp.NewContext(context.Background())
defer cancel()
if err := chromedp.Do(ctx); err != nil { // starts the browser
log.Fatal(err)
}
var wg sync.WaitGroup
for _, url := range urls {
wg.Go(func() {
tabCtx, cancel := chromedp.NewContext(ctx) // a new tab
defer cancel()
if err := chromedp.Do(tabCtx, chromedp.Navigate(url)); err != nil {
log.Println(url, err)
}
})
}
wg.Wait()


> I cannot reach an element or the network events of an iframe

Chrome runs an iframe from another site (a cross-site iframe) in its own
process, as a separate target of the type `iframe`. The DOM tree of the page does
not hold its content. The node of the `iframe` element has no `ContentDocument`,
so a query with `FromNode` finds nothing and waits until the context ends.
`chromedp` does not attach to the iframe target by itself, so the network events
of the iframe do not reach the context of the page. An iframe from the same site
is not affected.

To work in the iframe, find its target with `chromedp.Targets` and attach to it
with `chromedp.WithTargetID`. The new context runs actions in the iframe and
gets its events. The parent context must have run once, so that it has a
browser.```go
infos, err := chromedp.Targets(ctx)
if err != nil {
    log.Fatal(err)
}
for _, info := range infos {
    if info.Type == "iframe" && strings.HasPrefix(info.URL, "https://other.example/") {
        frameCtx, cancel := chromedp.NewContext(ctx, chromedp.WithTargetID(info.TargetID))
        defer cancel()
        text, err := chromedp.Run(frameCtx, chromedp.Text(chromedp.CSS("#inner")))
        // ...
    }
}

The other way is to turn off site isolation, so that the iframe stays in the
process of the page. Add Flag("disable-features", "SitePerProcess,IsolateOrigins")
and Flag("disable-site-isolation-trials", true) to the options of the exec
allocator. Then FromNode and the events of the page reach the iframe. This
turns off a security feature of Chrome. Use it only for pages that you trust. See
docs/decisions/2026-10-04-keep-site-isolation-on.md.

> SendKeys with a new line inserts two line breaks

SendKeys sends the character \n as the Enter key, in the form \r. That is a
keyDown event, a char event with the text \r, and a keyUp event. The key
kb.Enter is the same as \r, so it sends the same events and does not help. An
editor that handles Enter on keydown and cancels the default, such as Lexical,
can insert two line breaks, because the char event inserts one more. To send the
Enter key with no char event, send a rawKeyDown event and a keyUp event with
input.DispatchKeyEvent, and send the text before and after it with SendKeys:```go
enter := chromedp.Func(func(ctx context.Context, t *chromedp.Target) error {
for , typ := range []input.DispatchKeyEventType{
input.DispatchKeyEventTypeRawKeyDown, input.DispatchKeyEventTypeKeyUp,
} {
, err := cdp.Call(ctx, t, input.DispatchKeyEvent, input.DispatchKeyEventParams{
Type: typ, Key: "Enter", Code: "Enter",
WindowsVirtualKeyCode: 13, NativeVirtualKeyCode: 13,
})
if err != nil {
return err
}
}
return nil
})
err := chromedp.Do(ctx,
chromedp.SendKeys(sel, "Hello"),
enter,
chromedp.SendKeys(sel, "World"),
)


> The browser closes when the timeout of my context ends

The first `Run` on a context starts the browser, and it binds the life of the
browser to the context that you pass to that call. When that context ends, the
browser stops. So a context from `context.WithTimeout` that you use for the
first `Run` closes the browser when the timeout ends, and every later call on the
parent context fails with `context canceled`.

To limit one action, start the browser first with a context that has no timeout.
Then run the action with a context that you derive from it. When the timeout
ends, `Run` returns an error that wraps `context.DeadlineExceeded`. The tab and
the browser stay open, and `ctx` still works. `Run` has no option for the timeout
of one action.```go
ctx, cancel := chromedp.NewContext(context.Background())
defer cancel()
if err := chromedp.Do(ctx); err != nil { // starts the browser, with no timeout
    log.Fatal(err)
}

tctx, tcancel := context.WithTimeout(ctx, 5*time.Second)
defer tcancel()
err := chromedp.Do(tctx, chromedp.Navigate(url)) // only this call has the timeout

> Why does chromedp pass disable-dev-shm-usage, and can I turn it off?

DefaultExecAllocatorOptions sets disable-dev-shm-usage to true. Chrome then
keeps the files of its shared memory in the temporary directory, such as /tmp,
and not in /dev/shm. The flag is on because /dev/shm is small in many
containers (64 MB in a default Docker container), and Chrome crashes when it
fills the space.

The cost is that the shared memory files are file-backed memory in the temporary
directory. They can be slower, and a program that runs for a long time and opens
many pages can see growing memory. If /dev/shm is large enough, for example a
memory volume that you mount in a container, turn the flag off after the default
options:go opts := append(chromedp.DefaultExecAllocatorOptions[:], chromedp.Flag("disable-dev-shm-usage", false), ) allocCtx, cancel := chromedp.NewExecAllocator(context.Background(), opts...) defer cancel()

> I want to use chromedp on a headless environment

Run the Go program that uses chromedp inside the
chromedp/headless-shell image. The image has
headless-shell, a smaller headless build of Chrome. chromedp finds it by
default.

Contributing

Read CONTRIBUTING.md before you send a change.
AGENTS.md holds the rules for people and coding agents. The tests
need Chrome or the headless-shell image. The repository holds two modules, and
AGENTS.md tells how to test each one.

These documents are in docs/:

Document Holds
docs/PLAN.md the purpose, the architecture and the open questions
docs/PROGRESS.md where the work stands
docs/BACKLOG.md known work that is not done
docs/API.md the new API, with old and new code side by side
docs/MIGRATION.md every renamed and removed name
docs/decisions/README.md the index of every recorded decision

Questions and ideas

Ask a question, or suggest a feature, in Discussions. The issue
tracker is for bugs. You can also chat on Discord.

Resources

  • headless-shell - A build of headless-shell that the tests use
  • chromedp: A New Way to Drive the Web - GopherCon SG 2017 talk
  • Chrome DevTools Protocol - Chrome DevTools Protocol reference
  • chromedp examples - More complicated examples for chromedp
  • github.com/chromedp/cdproto - Go reference for the generated Chrome DevTools Protocol API
  • github.com/chromedp/pdlgen - tool used to generate cdproto
  • github.com/chromedp/chromedp-proxy - a simple CDP proxy for logging CDP clients and browsers

常见问题

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

在客户端按下面的安装命令接入:/usr/local/bin/moling。装完在客户端配置里引用同一个包或端点。

Moling 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。

Moling 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。

Moling 需要 API Key 吗?

资料标注需要OAuth 授权认证,使用前准备好对应凭据,并按最小权限配置。

Moling 支持哪些客户端?

资料标注支持 Claude Desktop、Claude、Cursor、Cline、Codex、Zed;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。

Moling 能提供哪些能力?

资料显示能力标签包括 agent、ai、mcp、ai assistant、ai agent、llm。具体工具清单以官方文档为准。

Moling 需要什么运行环境?

运行时为 go。

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。