Devices & access
Sign in with your BrowzWork account to manage your devices. This page cannot read files or run commands.
Devices
Machines available to your AI clients through Remote MCP.
Your devices
Only devices linked to this accountWhere you use it
Web MCP
Manual custom connector in Developer mode; not a public plugin.
Custom Remote MCP connector.
Custom Remote MCP connector.
Others
Remote HTTP MCP for Cursor Cloud Agents.
Use the shared Remote HTTP MCP endpoint. OAuth and client compatibility require separate verification; do not reuse another client's identity.
Harness MCP
DSH Desktop is the PC form of DSH. Both use the same managed devices, added Workspaces, and DSH account identity.
Shared endpoint: https://mcp.browz.work/mcp. Each client must complete its own supported OAuth flow.
Connect ChatGPT
Create a BrowzWork custom MCP app in ChatGPT web.
ChatGPT custom MCP apps use Developer mode. Full write/modify MCP support is currently available on Business and Enterprise/Edu; Pro custom MCP support is more limited. Availability also depends on workspace role and policy. BrowzWork does not change those eligibility rules.
1 Prepare your Agent
Install the local Agent and approve the device in BrowzWork. The Default Workspace, Workspace Access, and background connection are configured automatically.
2 Create the app
In ChatGPT web, enable Developer mode. On Business, admins/owners create custom apps; on Enterprise/Edu, authorized developers can also create them after RBAC access is granted. Use Settings → Apps → Create or Workspace settings → Apps → Create as available, name the app BrowzWork, and enter the MCP endpoint below.
https://mcp.browz.work/mcp3 Authenticate and scan tools
Choose OAuth authentication. If ChatGPT exposes a custom OAuth client field, use the public client ID below and no client secret. Complete BrowzWork sign-in, then click Scan Tools.
browzwork-chatgptKeep the server-discovered OAuth endpoints, MCP resource, and scopes unchanged. Do not paste any BrowzWork password or token into the app configuration.
4 Create and test
After the tool scan succeeds, click Create. Start a new chat, enable or @mention BrowzWork, then ask for device_list, select your device and Workspace, open a session, and read a harmless test file.
Write and command tools follow the locally granted Access Scope and Agent safety policy.
ChatGPT setup: OpenAI's current Developer mode guide ↗. Never paste passwords, tokens, private keys, or pairing proofs into ChatGPT.
Connect Claude
Custom connector setup using your BrowzWork account permission.
Claude remote custom connectors are available on Free, Pro, Max, Team, and Enterprise plans; Free accounts are limited to one custom connector. Complete BrowzWork OAuth sign-in, then verify device selection, session open, and a harmless read. Sign in with the account that owns the device to use its added Workspaces.
1 Prepare the Agent
Install the Agent and approve the device in BrowzWork. The Default Workspace, Workspace Access, and background connection are configured automatically.
2 Add a custom connector
For individual Free/Pro/Max accounts, open Customize → Connectors → + Add → Add custom connector. For Team, an Owner or Primary Owner uses Organization settings → Connectors → Add → Custom → Web; Enterprise also permits authorized custom roles with connector-management access. Name it BrowzWork and enter this server URL.
https://mcp.browz.work/mcp3 Choose OAuth
Select Sign in now and Use your own OAuth client. Enter this public client ID and leave the secret blank.
browzwork-claudeKeep the discovered endpoints and mcp:tools scope. Do not choose Claude's published identity or automatic registration for this server.
4 Connect and test
Finish adding the connector and complete BrowzWork sign-in. Then enable BrowzWork in a Claude conversation and test device list, selection, session open, and a harmless file read before edits or commands.
If OAuth sign-in cannot return to Claude, do not change BrowzWork endpoints manually; contact support with the error shown by Claude.
Claude setup: Anthropic's current custom connector guide ↗.
Connect Grok
Custom connector setup using your BrowzWork account permission.
Custom connector support is in beta. Complete BrowzWork OAuth sign-in, then verify device selection, session open, and a harmless read. Sign in with the account that owns the device to use its added Workspaces.
1 Prepare the Agent
Install the Agent and approve the device in BrowzWork. The Default Workspace, Workspace Access, and background connection are configured automatically.
2 Add the custom MCP connector
Open grok.com/connectors, choose New Connector → Custom, and enter this public MCP server URL.
https://mcp.browz.work/mcp3 Complete authentication
Let Grok discover the server authentication and complete BrowzWork sign-in. If Grok exposes advanced OAuth client fields, use the dedicated public client ID below with no client secret.
browzwork-grokDo not manually enter authorization or token endpoints unless Grok explicitly asks for them; use the server-discovered metadata.
4 Connect and test
Finish the connector flow and complete BrowzWork sign-in. In a Grok conversation, enable BrowzWork and test device_list, device selection, session open, and a harmless file read before editing or running commands.
If OAuth sign-in fails, keep the server URL unchanged and use the error shown by Grok for troubleshooting.
Grok setup: xAI's current connector guide ↗.
Connect Cursor Web
Use BrowzWork from Cursor Cloud Agents through the shared Remote HTTP MCP endpoint.
Cursor Web is a non-standard compatibility entry under Others, alongside other MCP-compatible clients. ChatGPT, Claude, and Grok are the standard Web MCP entries. Cursor is not part of the 0.1.17 release gate. It still uses the same devices, added Workspaces, Access Scope, Gateway, and Native Agent; it does not install a Cursor-specific Agent.
1 Prepare the Agent
Confirm the device is paired and online. In Local Settings, explicitly update account access for the intended Workspace so the Web MCP bindings, including Cursor, and the DSH Harness MCP binding are present.
2 Add the remote MCP server
In Cursor Web or Cloud Agents, open the MCP menu. Team administrators can also use Dashboard → Plugins & MCPs for a shared server. Add a custom HTTP server named BrowzWork with this URL.
https://mcp.browz.work/mcp3 Authenticate
Configure static OAuth with the client ID below, no client secret, and scopes openid mcp:tools offline_access. Complete sign-in with your BrowzWork account. Use this server configuration where Cursor accepts MCP JSON:
{
"mcpServers": {
"BrowzWork": {
"url": "https://mcp.browz.work/mcp",
"auth": {
"CLIENT_ID": "browzwork-cursor",
"scopes": ["openid", "mcp:tools", "offline_access"]
}
}
}
}browzwork-cursorThe registered Web callback is https://www.cursor.com/agents/mcp/oauth/callback. Do not use a Desktop localhost callback, a client secret, SSE, or mcp-remote for this Web setup.
If your Cursor form only supports dynamic registration and cannot preserve the static client ID, that form is not yet verified. Do not substitute another channel's client ID. Cursor static OAuth instructions ↗
4 Verify the managed path
In a new Cursor Web agent, list devices, select the intended device and Workspace, open a session, list the Workspace root, and read a harmless test file. Then verify one bounded write only if the local Access Scope permits it.
OAuth success alone is not a full gate pass; the real Cursor callback, token claim, tool discovery, and one remote file operation must all be recorded.
Cursor documents remote Streamable HTTP MCP and per-user OAuth for Cloud Agents. BrowzWork keeps Cursor as a distinct browzwork-cursor Principal rather than sharing another Web MCP identity. If the device is online but its Workspace is not visible, check device ownership, the installed Agent version and the added Workspace list. Owner bindings are repaired automatically; no separate Workspace grant is required. This guide does not certify the pending four-platform real-client release tests.
Connect DSH Desktop
Install the BrowzWork plugin and use managed Workspaces across your devices.
This is a development prerelease for DSH Desktop 0.2.0-rc.2. It does not replace the DSH native Workspace: BrowzWork tools always stay inside the selected managed Workspace and never fall back to a local path.
1 Prepare the Agent
Install the BrowzWork Agent on each computer you want to use. Confirm that the device is online, add the required Workspace in Local Settings, and leave remote access enabled.
2 Install the plugin
Download browzwork-dsh-bundle-0.1.0-alpha.7.tgz. Its SHA-256 is b59e1305881c8eef4a0fabe7242ad04d9b2f13efcc9d6e7bbb8936c0731a010f. Close DSH Desktop, then install it with the official DSH plugin manager and restart DSH Desktop completely.
dsh plugin --profile desktop add <downloaded .tgz>Requires Agent 0.1.17. The DSH Desktop plugin remains a development prerelease. Execution-context history and hidden healthy status-sync rows on stock Desktop 0.2.0-rc.2 also require the reviewed core SDK compatibility update; plugin installation alone does not apply it. Use the official plugin manager for the plugin package.
3 Sign in and select your device
Open the BrowzWork sidebar in DSH Desktop and complete BrowzWork sign-in. Select your device and an added Workspace. The Device Owner can use all Workspaces added to the device through browzwork-dsh; no separate Workspace authorization is required. Use Local Settings ↗ to add or remove folders.
If the sign-in page asks you to return and refresh, use the BrowzWork sidebar's Refresh action. If it is not yet visible, close and reopen the BrowzWork sidebar; a full DSH restart is the recovery path for this prerelease.
4 Select and verify
Select a device and one of its added Workspaces, then test device_list, workspace_list, session_open, fs_list, and a harmless fs_read.
The plugin remembers the selected device and Workspace by their visible names. New conversations silently revalidate the saved choice and reuse a valid managed session; expired sessions or offline devices return a specific error code and recovery message.
The plugin release remains a prerelease until the twelve Windows x64/Linux x64/Linux arm64/macOS arm64 standard Web MCP golden paths (ChatGPT, Claude, Grok) and the two Windows/macOS DSH Desktop Harness MCP paths pass the release gate. Cursor Web and other MCP-compatible clients are non-standard Others, not release blockers.
Install the Agent
One command on a computer you control.
Connect this computer
Node.js 20+ and npm requiredOpen PowerShell on Windows or a terminal on Linux or macOS, then run:
npx --yes @browzwork/agent@0.1.17The command verifies the native Agent, creates a local Default Workspace on first use, and opens browser verification. Sign in, compare the short code, and approve this device. Default is added and immediately available to the Device Owner with Workspace Access. After new pairing the Agent enables remote connections and continues running in the background automatically.
Normal setup needs no account access code. Browser verification identifies your BrowzWork account and the Agent configures the Default Workspace directly.
Legacy migration
Only use this if an updated Agent explicitly asks for an account access code. The code identifies your BrowzWork account only; it is not tied to ChatGPT, Claude, or Grok and is not a password or access token.
Agent 0.1.17 supports Windows x64, macOS Apple Silicon, Linux x64, and Linux ARM64.
Next steps
After browser verification, BrowzWork is ready to use with the Default Workspace and the Agent remains running in the background even after the terminal closes. Use Open local Agent on the Devices page when you want to add another Workspace or open Advanced settings. Headless machines can still use workspace add.
Usage
Account activity and call trends.
Remote calls
— / 10,000Data transfer
— / 500 MBDaily average
—Calls over time
Last six calendar months, counted from validated Remote MCP tool calls.
Loading usage…
Plan & billing
View your plan and upgrade securely.
Free Current plan
$0 / month
10,000 Remote Calls / month
500 MB Data Transfer / month
3 Devices
Web MCP: ChatGPT, Claude, Grok; Others: Cursor Web and other MCP-compatible clients. Harness MCP: DSH.
Pro Current plan
$15 / month
50,000 Remote Calls / month
5 GB Data Transfer / month
15 Devices
Secure checkout opens on mcp.browz.work. Pro starts only after Paddle confirms your subscription to BrowzWork.
Online checkout is not available yet. Ask support about Pro.
Team
$20 / seat / month
50,000 Remote Calls × seats / month, shared pool
5 GB Data Transfer × seats / month, shared pool
100 Devices × seats, shared pool
Team is inquiry-only in this release. There is no Team checkout or seat billing.
Email usBilling records
Invoices are not available in this dashboard yet. For billing help, contact support@browz.work.
Settings
Account and appearance.
Signed-in account
Loading account…
Account & security
Manage your email, password, and sign-in security in the BrowzWork Account Console.
Appearance
This setting affects the dashboard in this browser only.
Sign-in session
Signing out clears only this page's temporary token. It does not revoke devices or other app sessions.
Permission boundary
The cloud shows device names, ownership, and online status. Workspace paths, file contents, and command history stay on your device.
File, search, edit, and command permissions remain under the local Agent policy. The Device Owner can use every added Workspace. OS Access remains an explicit device-level setting.