Google Discover MCP Server and API by Cotlet

How to work in an app by clicking in the user's Cloud Browser with Cotlet's browser tools: which browser to use for this workspace and the sign-in link, the method step by step, the app's screens, menus and actions, notes, and the app's dedicated tools when Cotlet has them. Call it before the first browser step in an app. Works for any web app; one outside Cotlet's catalog gets the general method. Free, no browser session. For Google Discover it covers these actions: Read Discover feed, Check domain in feed.

The apps of the user's workspace and whether each works now: the app account in the user's Cloud Browser (connected, signed out, signed in but not connected), and the one step the user has to take, with the console link. Free. Needs the user's Cotlet API key or sign-in.

Clicks element `element`, or point `x`, `y` of the screenshot; `double` double-clicks, `button: right` right-clicks. Answers the page after it: the elements and text in view, with the same numbers. Says when nothing changed. A link that would open a new tab opens in this one. Buttons that pay or buy, and sign-in and account pages, are refused.

Ends the browser session when the task is done: the browser keeps its cookies and sleeps, and the per-minute charge stops. Answers the session's minutes and credits.

Gets the file that clicking `element` (a download link or button) or opening `url` starts: answers its name, type, size and a link valid 15 minutes; CSV, JSON and other text files also as text.

Drags from one element or point to another: a card between columns, a piece on a board, a slider's handle. Answers the page after it: the elements and text in view, with the same numbers.

Runs a JavaScript function body on the page and answers what it returns, as JSON. Use it to extract data the page holds (a table, a list, attributes); act with the other tools. It cannot send requests, click, submit, or read cookies or storage; a filled field's value reads [hidden by Cotlet]. Example: return [...document.querySelectorAll('h2')].map(h => h.innerText)

Fills several fields in one step: text fields take text, drop-downs an option's label, checkboxes, radios and switches true or false. Stops at the first field that fails and names those filled. Does not send the form: click its button after. Answers the page after it: the elements and text in view, with the same numbers.

Finds `text` on the whole current page, frames included: the elements whose label holds it, with their numbers, and each line of text that holds it. Use it instead of reading or scrolling a long page for one thing.

Rests the pointer on an element or a point, to open a hover menu, a tooltip or a preview. Answers the page after it: the elements and text in view, with the same numbers.

Opens `url` in the current tab, or goes `history` back or forward. Answers the page after it: the elements and text in view, with the same numbers.

Starts the user's Google Discover Cloud Browser (a real Chrome on Cotlet, signed in to Google Discover as the user; use it for what the Google Discover tools do not cover) and opens `url`. Call it first. Only Google Discover pages open. Answers the page after it: the elements and text in view, with the same numbers. In Google Discover it serves these actions: Read Discover feed, Check domain in feed.

Presses a key on the focused element: Enter to send, Tab to the next field, Escape to close a pop-up, arrows in a list or slider, or one character (a site's shortcut). Answers the page after it: the elements and text in view, with the same numbers.

The whole current page: its text (up to 20,000 characters) and a numbered list of its links, buttons, fields and other controls, open dialogs first. Use it on a new page or when a step's answer is not enough; `element` reads only inside one element (a dialog, a form, a list). Never returns field values, cookies or storage.

A JPEG of the visible part of the page. Use it for boards, canvases, maps, charts and layout that the text does not show, then click or drag by `x`, `y` of the image.

Scrolls with the mouse wheel: `down` or `up` by `screens`, to `top` or `bottom`, until `element` is in view, or inside a panel or list (`inside`). Content that loads on scroll appears. Answers the page after it: the elements and text in view, with the same numbers.

Chooses `option` (its label as listed, or its value) in drop-down `element`. For a home-made drop-down, click it, then click the option. Answers the page after it: the elements and text in view, with the same numbers.

Types `text` into field `element`, replacing what it held; `submit` presses Enter after it. For several fields use fill_form. Without `element` the text goes where the focus is, as a person types after clicking: for spreadsheet cells and editors that read does not number (Google Sheets, Docs), click the place first; there a tab in `text` presses Tab and a line break Enter, so one call fills a table. Answers the page after it: the elements and text in view, with the same numbers. Password, card and one-time code fields are refused: the user types them.

Puts files into file field or upload button `element`: each file is {"url": "https://..."} or {"file_id": "..."} of a file uploaded to Cotlet; up to 10 files, 20 MB each. Answers the page after it: the elements and text in view, with the same numbers.

Waits until `text` shows on the page (at most `seconds`, default 5, up to 20), or just `seconds`: for results that load slowly. Says whether the text appeared. Answers the page after it: the elements and text in view, with the same numbers.

Status of an earlier Cotlet call, by its call_id: pending (waits for the owner's approval in the Cotlet console), approved, running, success, rejected, expired, blocked or failed. For a call that answered with status pending or running.

Sends a report to the Cotlet team: a tool that failed or returned wrong data, an error you could not get past, or a step the browser could not do, with the page, the step and what you saw. A missing dedicated tool is not a problem to report: do the task in the browser. A feature the user asks for goes as feature_request. A report promises nothing: tell the user it was sent, never that something will be built or fixed, or when. The report carries the error text and call_id when there is one, never passwords or tokens. Needs the user's Cotlet API key or sign-in.

Requests access to apps for the user's workspace, as Request access in the console: every app is granted at once. Not needed to work in an app: the user's AI works in any web app in their Web browser with cotlet_app_guide. An app with dedicated tools gets them after the user signs in to it in their Cloud Browser. Access says nothing about whether a task works there: check that in the browser. Takes app slugs as listed at https://cotlet.dev/api/v1/apps. Free. Needs the user's Cotlet API key or sign-in.

Returns a one-time upload_url and a file_id for a local file (an image for a post, an attachment). The file is sent to upload_url with an HTTP PUT; an argument that takes a file accepts {"file_id": "..."}. Up to 25 MB, the URL works once. Free. Needs the user's Cotlet API key or sign-in.

The Cotlet account this connection acts as: email, workspace, access status, the API key or OAuth client in use with its scope, the apps the workspace may call and its credit balance. Free. Needs the user's Cotlet API key or sign-in.

Check domain in feed in the user's own Google Discover account from ChatGPT, Claude or any AI agent. Returns how to do it in Google Discover step by step (screens, buttons, fields) and opens nothing by itself; then do it with the cotlet_browser_* tools in the user's Google Discover Cloud Browser. Google Discover's official API: Google Discover has no API for the feed (checked 25 Oct 2026).

Read Discover feed in the user's own Google Discover account from ChatGPT, Claude or any AI agent. Returns how to do it in Google Discover step by step (screens, buttons, fields) and opens nothing by itself; then do it with the cotlet_browser_* tools in the user's Google Discover Cloud Browser. Google Discover's official API: Google Discover has no API for the feed (checked 25 Oct 2026).

Up to 300 different cards Discover shows in a country, from many reads, with how often each appeared. One read of the Discover feed holds about 31 cards, and each read draws them from a larger pool. This tool reads the feed of the country many times in parallel, each time from another home IP there, until it has the number of different cards asked for or the pool runs dry, and returns them at once. seen_in says in how many reads a card appeared (a card in most reads is the core of the feed, one seen once is rare) and best_position its highest place. Cards come most often seen first. complete is false when the feed held fewer different cards than asked for in the time of one call. A call takes up to about 70 seconds and costs more credits than one read, because it makes many.

The cards Google Discover shows right now in a country, in feed order, with no Google account. Returns the Discover feed Chrome for Android shows on its New Tab page to someone who is not signed in to Google, read from a home IP in the country asked for: position, title, publisher, URL, domain, article or video, age as the card shows it, and the card image. It is the default feed of that country and language, not anyone's personal feed and not data from a panel of users. One read holds about 30 cards; the feed changes over the day, and a read is reused for up to 15 minutes per country and language.

Whether cards from a domain are in a country's Discover feed now, and at which positions. Reads the same feed as get_discover_feed (signed out, from a home IP in the country) and keeps the cards whose link is on the domain or one of its subdomains. in_feed is false when none is there right now; feed_size says how many cards the feed held. Run it on a schedule to see when a story enters and leaves.