Tool reference
Every tool your agent can call, documented one by one.
35 tools, grouped the way the MCP server registers them. Each page carries the exact call, its parameters, the response shape, what it costs, and when a neighbouring tool is the better pick.
Three ways in
One tool set, three call surfaces.
The same tools are reachable from an MCP client, from the REST API, and from a plain cURL. Each tool page prints the MCP invocation and the REST equivalent side by side.
8 tools
Core — run, compile, replay, dispatch
inspect_siteAsk what a site did last time, before you pay to find out again
Decide how to run against a host before the run, using evidence the platform already has.
run_goalRun a named goal on a live page
Do one thing on one site, right now, and block until you know whether it worked.
compile_skillSolve a task once, keep the minimal path
Turn a flow you are going to run many times into a stored artifact that costs almost nothing to run again.
run_skillReplay a compiled skill with no model in the loop
Run a known-good flow again, as cheaply as it can possibly be run.
dispatchMatch a re-phrased goal to a skill you already have
Be the one call an agent makes, and get cheaper every time the agent asks for the same thing in different words.
list_skillsList the compiled skills you can replay
Answer “what do I already have?” before spending anything.
list_catalogRead the run contract before you spend a credit
Let an agent plan: pick a skill it can actually run, with values it can actually supply.
search_librarySearch the shared corpus of compiled flows
Find out whether somebody has already solved this shape of task on this host.
3 tools
Async jobs
submit_runStart a run and get the id back immediately
Run something that will outlive your request timeout, or run twenty things at once.
get_jobPoll an async job for its status and result
Find out whether the thing you started has finished, and get what it produced.
cancel_jobStop a running job and get the credits back
Change your mind about a job that is still running, without paying for it.
10 tools
Observation & data
observe_pageSerialize a page into an indexed element map
Let an agent see what is on a page before deciding what to do about it.
screenshotCapture a page once it has finished painting
Get one honest image of a page, without catching it mid fade-in.
download_assetGet the actual file, not just a reference to it
Turn a URL you already have into the bytes it points at.
extractRead a page into the JSON shape you asked for
Get data off a page in the shape your code already expects.
searchRanked web results, optionally with the pages read
Find pages, and optionally have them read for you, in one call.
deep_searchSearch, then actually read the results
Do the research pass an agent would otherwise fake from search snippets.
map_siteEnumerate a site’s URLs without reading them
Find out what is on a site before deciding how much of it to read.
submit_crawlWalk a whole site, bounded and resumable
Read a defined slice of a site, once, without either missing it or over-reading it.
etlIngest a source into a queryable content store
Turn web content into something you can ask questions of, without building a pipeline.
etl_querySemantic search over what you ingested
Ask a question of the content you ingested, and get back the passages that answer it.
4 tools
Monitors
create_monitorWatch a page and get a signed webhook when it changes
Know when something on a page changed, without polling it yourself.
list_monitorsList the monitors this tenant is running
See what you are already watching, and what it is costing you per hour.
delete_monitorDelete a monitor and stop the recurring charge
Turn off a watch you no longer need.
get_monitor_historySee every check, not just the ones that fired
Tell the difference between “nothing changed” and “the check has been broken for a week”.
8 tools
Sessions, credentials & HITL
connect_accountSend a link, let the person sign in themselves
Get logged into a site that an agent is not going to be allowed to log into.
list_connectionsCheck whether the sign-in you sent was completed
Stop an agent from re-running a task before the user has actually signed in.
list_accountsWhich logins you hold, masked
Answer “can I get into this site, and as whom?” before starting a run.
list_warningsFind out which account or session is broken
Stop retrying a task blind when the real problem is a broken login, not the goal.
acknowledge_warningClear a warning once you have dealt with it
Close the loop after fixing what a warning named — a missing secret now stored, or a fresh session captured.
control_runDrive a paused run’s browser yourself
Clear a blocker you can see on the page, when no code and no credential would have helped.
submit_verificationHand a code back to a run that is waiting
Finish a sign-in that got as far as the second factor.
solve_captchaHand a CAPTCHA to the solver and continue
Clear the one wall that is not a code, not a login, and not a click.
Start with one call.
Point an MCP client at the server, or send a Bearer-authed POST. Both reach the same engine.