Introduction

The DSH plugin ecosystem emphasizes “everything is a plugin,” splitting installable, overridable, and removable capabilities into bundles; the community directory is an independent site with no official affiliation to DeepSeek / Huafang and should not be interpreted as an official app store.

In agent automation, a common pain point is: the model not only needs to “understand web pages” but also be able to operate the user’s currently logged-in, context-aware real browser. Traditional headless browsers or external scraping pipelines often lack login state and are difficult to cover actions like local file upload, screenshot, and PDF export.

Below introduces dsh-kimi-webbridge. It exposes the local Kimi WebBridge daemon to the DeepSeek Harness (dsh) agent, enabling the agent to drive the current user’s real browser via a local daemon.

What is this

dsh-kimi-webbridge is a DeepSeek Harness plugin maintained by MicroHEROX, licensed under MIT.

It does the following specific things:

  • The agent side obtains a set of kimi_webbridge_* tools.
  • The plugin collaborates with the browser side via a local daemon.
  • The target is the current user’s real browser.
  • The flow stays on the machine: plugin -> 127.0.0.1:10086 daemon -> user's real browser.
  • It does not rely on third-party services to relay browser traffic.
  • No API keys are required.
  • Requires Node.js >= 18.

What it solves is not “building another browser sandbox,” but enabling dsh agents to complete page navigation, reading, clicking, filling, screenshotting, network viewing, uploading, and PDF exporting within the user’s logged-in browser environment.

Core Capabilities

The plugin provides 15 tools, named as follows:

kimi_webbridge_navigate
kimi_webbridge_find_tab
kimi_webbridge_list_tabs
kimi_webbridge_snapshot
kimi_webbridge_click
kimi_webbridge_fill
kimi_webbridge_evaluate
kimi_webbridge_cdp
kimi_webbridge_screenshot
kimi_webbridge_network
kimi_webbridge_upload
kimi_webbridge_save_as_pdf
kimi_webbridge_close_tab
kimi_webbridge_close_session
kimi_webbridge_start_daemon

Covered capabilities include:

  1. Open page.
  2. Find or list tabs.
  3. Read page snapshot.
  4. Click element.
  5. Fill form.
  6. Execute JavaScript in page.
  7. Screenshot.
  8. View network requests.
  9. Upload file.
  10. Save current page as PDF.
  11. Close tab.
  12. Close entire session.
  13. Attempt to start local daemon when daemon is unreachable.

All these tools are prefixed with kimi_webbridge_, making them easy to identify and filter within the dsh tools directory.

Installation and Activation

First use the official installation command. Here taking the demo profile as an example:

dsh plugin --profile demo add github:MicroHEROX/dsh-Kimi-WebBridge

After installation, start the profile with web capabilities:

dsh --profile demo web

After activation, kimi_webbridge_* tools will appear in the agent directory.

If you need to run in a one-time overlay manner, you can use:

dsh web --patch ./kimi-webbridge.overlay.yml

This command is only the entry point; the specific overlay content should be placed in the existing kimi-webbridge.overlay.yml on the local machine.

To remove the plugin, use:

dsh plugin --profile demo remove dsh-kimi-webbridge

Typical Usage

Below is a reproducible prompt example.

First install and start:

dsh plugin --profile demo add github:MicroHEROX/dsh-Kimi-WebBridge
dsh --profile demo web

Then give the agent a task like this:

Open example.com in the browser, tell me what's on the page, and screenshot it.

After the above steps, the agent is expected to complete these things:

  1. Open target page.
  2. Read page content.
  3. Execute screenshot.
  4. Return the result to the user.

If you need to temporarily enable without changing the default installation method, you can also use the overlay command:

dsh web --patch ./kimi-webbridge.overlay.yml

Applicable Scenarios and Notes

Suitable scenarios:

  • Need to operate the user’s currently logged-in browser.
  • Need to read real pages, take screenshots, and save PDFs.
  • Need to complete web automation on the local machine rather than handing page traffic to third-party services.
  • Need to incorporate browser operations into the dsh agent’s toolchain.

Notes on limitations:

  1. fill / click might be ignored by sites that strictly check event.isTrusted, such as some bank portals or CAPTCHA pages. These pages may require manual intervention.
  2. Cross-origin iframes are not within the current scope; relevant tools primarily operate on the top-level frame.
  3. Sessions are managed by profile dimension, not isolated by agent dimension; subagents will share the same set of tabs.
  4. Windows has been tested; macOS / Linux are supported by code paths but not yet verified.
  5. All configuration keys are optional.
  6. The plugin runs with the permissions of the current dsh process; you should check the source code and license before installing.

There is also a known issue on the dsh harness side: in rare transient pnpm failure cases, dsh-kimi-webbridge may remain in dsh.profile.bundles, causing profile startup failure. The recovery method is to edit the corresponding profile’s package.json, remove the relevant entries, and then start.

Links

GitHub repository:

https://github.com/MicroHEROX/dsh-Kimi-WebBridge

Community directory: The factual basis for this article does not provide a specific directory page URL, so no unverified links are added.