Actions

Actions

Every action is an MCP tool and an HTTP endpoint with the same parameters, and every read is an MCP resource and an HTTP endpoint. A test fails the build if one exists without the other. A client that cannot read resources reads the same URIs with read_resource.

Browser actions

Everything you can do to a page.

MCP	Endpoint	Does
open_session	POST /browser	Open this session's browser, or pick up the one it was using.
session://current	GET /browser	What this session is, and whether it holds a browser.
navigate	POST /browser/navigate	Go to a URL. Returns the URL and title the browser ended on.
interact	POST /browser/interact/{action}	A mouse action on an element: click, double_click, right_click, hover or scroll_to.
drag	POST /browser/drag	Drag an element onto another element (to), or by an offset in pixels (by_x, by_y); give one or the other. A range slider is by_x; a card into a column is to.
write	POST /browser/write	Type into an input, textarea or contenteditable, and read the value back so
press_key	POST /browser/press-key	Press a key or a combination, at an element (selector) or wherever focus already is.
extract	POST /browser/extract	Read an element's visible text and HTML.
assert	POST /browser/assert	Check the page with JavaScript that must return true; otherwise the call fails. Return a boolean, not the thing itself: return !!document.querySelector('#total').
outline	POST /browser/outline	Map the page: the elements worth acting on, each with a selector checked to match exactly one element, and whether it can be used now.
screenshot	POST /browser/screenshot	Capture a PNG of the viewport, of one element (selector), or of the whole
print	POST /browser/print	Print the page into the session's files, kept past the browser.
execute_script	POST /browser/script	Run JavaScript in the page and return what it returns.
frame	POST /browser/frame	Move into an iframe (switch, with a selector or an index), up one level (parent), or back to the page (default).
dialog	POST /browser/dialog	Answer a native alert, confirm or prompt: accept, dismiss, read (see the message without answering), or send_text (fill a prompt with text and accept it).
resize	POST /browser/resize	Resize the browser window.
upload_file	POST /browser/upload	Attach a file to a file input (selector). Give exactly one source:
end_browser	DELETE /browser	Quit the browser, keeping the session and its context.

Flows

A sequence of the actions above, saved under a name and run in one call. See Flows for what a flow document looks like.

MCP	Endpoint	Does
flow://flows	GET /flows	Every flow this session can run.
flow://flows/{name}	GET /flows/{name}	One saved flow, with its steps.
flow://schema	GET /schemas/flow	The shape of a flow document.
save_flow	PUT /flows/{name}	Create or replace a flow.
run_flow	POST /flows/{name}/runs	Run a saved flow.
delete_flow	DELETE /flows/{name}	Delete one of this session's flows.

Files

What a session has produced, and how to keep one past the browser that made it. See Files.

MCP	Endpoint	Does
session://files	GET /files	Every file this session has.
keep_file	PUT /files/{name}/kept	Keep one download beyond its browser.

Secrets

The secrets a caller may type into a field without ever seeing them. See Secrets.

MCP	Endpoint	Does
secret://secrets	GET /secrets	The secrets this caller may bind.

Every call names its session — an X-Session-Key header or ?session= on the URL. There is no session id anywhere; see Sessions.

---

Installing · Deployment · Administration