MCP Tool Overview
DevBridge provides 25 MCP tools across 9 categories.
Page Reading
| Tool | Description |
|---|---|
read_page | Get the accessibility tree of the current page — elements, refs, and structure |
get_page_text | Extract the main text content of the page |
find | Find elements on the page by text, role, or selector |
Interaction
| Tool | Description |
|---|---|
computer | Mouse and keyboard actions — see sub-actions below |
form_input | Fill form inputs with React-compatible value injection, supports contenteditable |
upload_image | Upload an image to the active input |
file_upload | Upload a local file to a file input element |
computer sub-actions
Standard (13):
| Sub-action | Description |
|---|---|
left_click | Left-click on an element or coordinate |
right_click | Right-click on an element or coordinate |
double_click | Double-click on an element or coordinate |
triple_click | Triple-click (select all text in a field) |
hover | Move the mouse to an element without clicking |
type | Type text into the focused element |
key | Press a key or key chord (e.g. ctrl+z, Escape) |
scroll | Scroll at a position by delta |
scroll_to | Scroll an element into view |
left_click_drag | Click and drag from one coordinate to another |
zoom | Pinch-zoom gesture |
wait | Wait for a specified duration (ms) |
screenshot | Take a screenshot of the current page |
Canvas (12): See Canvas Interaction for the full list.
Navigation
| Tool | Description |
|---|---|
navigate | Navigate to a URL, go back, or go forward. Handles beforeunload dialogs |
Debugging
| Tool | Description |
|---|---|
javascript_exec | Execute JavaScript in the page context and return the result |
read_console | Read console logs from the current tab |
read_network | Read captured network requests; supports detail mode for full response bodies or bulk mode for summaries |
read_realtime | Read WebSocket frames and SSE events in real time (filter by time window or event count) |
Visual
| Tool | Description |
|---|---|
screenshot | Take a screenshot of the current page (also available as a computer sub-action) |
resize_window | Resize the browser window to specific dimensions |
GIF Recording
| Tool | Description |
|---|---|
gif_creator | Record a multi-step interaction as an animated GIF |
Tab Management
| Tool | Description |
|---|---|
tabs_context | Get information about all open tabs, including debugger attachment state and tracking start time |
tabs_create | Open a new tab |
tabs_close | Close a tab |
CDP / Extension Escape Hatch
When the standard tools don't cover your use case, you can reach directly into Chrome DevTools Protocol or call chrome.* APIs. See the Escape Hatch guide for details and examples.
| Tool | Description |
|---|---|
cdp_command | Send any CDP command directly (supports _help introspection to explore parameters) |
cdp_subscribe | Subscribe to CDP events; supports untilPredicate to stop automatically when a condition is met |
cdp_events_read | Pull historical CDP events from the ring buffer (pull mode — no persistent subscription needed) |
extension_call | Call whitelisted chrome.* API methods (e.g. chrome.storage, chrome.cookies) |
extension_subscribe | Subscribe to chrome.* events (e.g. chrome.tabs.onUpdated) |
extension_events_read | Pull historical chrome.* events from the ring buffer (pull mode) |
Other
| Tool | Description |
|---|---|
reset_session | Reset MCP session boundary: clear per-turn permission grants, hide visual indicators, drain in-flight requests, then reconnect. |
Every tool returns tab context
Every tool response includes tabContext with currentTabId, executedOnTabId, and availableTabs — so your AI always knows the current browser state.