DocuBench.AI

Product features

A browser workbench with controls built around the agent.

DocuBench gives MCP-compatible agents a visible desktop browser workspace, then surrounds that workspace with navigation boundaries, browser tools, site instructions, tab ownership, and reviewable activity.

Core capabilities

The control plane for browser-based agent work.

These features work together. Navigation control defines where the agent may go, browser tools define what it can do there, site instructions guide repeated work, and audit makes the session visible enough to review.

Boundaries

Navigation control

Set the browser workspace where an agent may operate. Known sites can open directly; unknown destinations can be blocked or sent back for approval.

Tools

Browser actions

Expose focused MCP tools for reading content, screenshots, clicking, typing, scrolling, dragging, and moving through browser history.

Review

Audit visibility

Watch activity during the session, and keep local audit records on MCP Power User when you need a saved trail after the work is done.

Memory

Site instructions

Attach site or URL-prefix notes so repeated workflows carry the same guidance, constraints, and operating preferences for AI agents.

MCP tab model

Three tab types for three kinds of browser work.

DocuBench separates general navigation, repeated site work, and developer-hosted workflows into distinct MCP-owned browser surfaces.

General scoped work

Navigation tab

The default MCP browser surface for research, operations, and one-off tasks inside the workbench's navigation policy.

MCP tab
allow or approve sites
read and interact
return tool results

Repeated site work

Site agent tab

A dedicated tab for a known website or URL scope, with tighter boundaries and reusable instructions for recurring workflows.

MCP tab
site-specific scope
site instructions
separate work context

Local UI workflows

Developer host tab

A controlled browser tab for localhost, staging, or another developer-owned app that an agent can inspect and operate.

MCP tab
localhost support
UI inspection
workflow testing

Browser tool set

Practical browser actions, exposed as structured MCP tools.

The agent does not receive arbitrary browser code execution. It asks DocuBench for focused actions: inspect the current page, capture visual state, interact with controls, and return a structured result.

tool surface
MCP
browser_list_tabs
browser_get_content
browser_screenshot
browser_click
browser_type
browser_scroll_tab
browser_history_step_back
browser_drag
Results stay reviewable

Content, screenshots, status, and errors come back as workbench results the user can inspect.

Activity review
local audit
TimeEventStatusInfo
12:02:13tool:navigateerrorNavigation denied by user
12:02:09tool:navigatependingReview unknown destination
11:59:56tool:screenshotokScreenshot returned
11:58:37connectokMCP client initialized

Audit and review

See the browser work as it happens, then inspect the trail.

Live review helps users understand current agent activity. Local audit persistence for MCP Power User keeps a saved record for people who need to examine browser actions after the session.

Live activity

Watch navigation and browser tool calls

Saved records

Keep local audit data on paid plans

Next step

Put a controlled browser workbench in front of your MCP agent.

Start with DocuBench Free, then upgrade when you need more concurrent work, longer sessions, multiple site agents, or local audit persistence.