WebMCP

Your agent can use Advibly the way you do.

Advibly puts its workspace on document.modelContext as 22 tools, offered as they become usable: a handful before you sign in, the whole workspace once you have. An agent running inside your browser can read your brand, draft into the composer and generate on-brand ads as you, on the session this tab already holds. No API key, no OAuth round trip, and nothing it does happens off-screen.

Live on this page

These are the real tools, right now.

The list below is read straight out of this tab's modelContext, not a hand-written table. Pick one and call it exactly the way an agent would.

CheckingReading document.modelContext…

Why in the page

The agent is a second pair of hands, not a second account.

No second identity

The tools run on the Clerk session and the live Convex subscription the page already holds. There is no token to mint, no consent screen, no per-agent API key - the agent is simply you, in your own tab.

One shared screen

A tool call is not a silent API request. The composer fills in with the prompt the agent wrote, the brand switches in the sidebar, and the render appears on the wall while you watch. You always know what it just did.

Approval before spend

Generations cost credits, so every one of them stops on an approval card showing the exact prompt, settings and price before anything is charged. Decline is a first-class answer the agent is told to respect.

Getting there

Three steps, no install.

  1. 01

    Open Advibly in an agentic browser

    ChatGPT's in-app browser ships WebMCP. In desktop Chrome, enable chrome://flags/#enable-webmcp-testing and restart. Everywhere else this page installs the WebMCP polyfill, so extension-based agents still find the tools.

  2. 02

    Sign in

    Signed out, an agent gets the public tools: what Advibly is, what it costs, which skills exist. Signed in with a brand onboarded, the whole workspace lights up.

  3. 03

    Ask for what you want

    "Look at my brand and make me three scroll-stopping image ads for the new launch, 9:16." The agent reads the brand, writes the prompts into the composer, and asks you to approve the spend.

The surface

Every tool, and what it does to your screen.

Available to everyone

Registered on every page, signed in or not. An agent can answer questions about Advibly from the page itself rather than from a scrape.

  • about_adviblyread-only

    What Advibly is, what it makes, who it is for.

  • get_advibly_overviewread-only

    The full markdown site overview (llms.txt).

  • get_pricingread-only

    Plans, prices and included credits.

  • list_agent_skillsread-only

    The installable Advibly agent skills and their artifacts.

  • get_mcp_connection_inforead-only

    How to reach Advibly from outside the browser (MCP + CLI).

  • open_signup

    Take the human to sign-up.

Once you are signed in

The agent reads the same workspace the human is looking at, and moves them around it. No API key, no OAuth handshake: it runs on the session the tab already holds.

  • get_page_contextread-only

    The current route, brand, composer state, selected product and credit balance.

  • list_brandsread-only

    Every brand on the account, and which is selected.

  • get_brandread-only

    A brand's identity: description, colours, tone, logo, kit.

  • select_brand

    Switch the active brand in the sidebar.

  • check_creditsread-only

    The credit balance.

  • list_generationsread-only

    Recent work with status and output URLs.

  • get_generationread-only

    Poll one generation until it finishes.

  • list_productsread-only

    A store brand's catalogue with photo URLs.

  • select_product

    Make one product the subject of what gets created next.

  • suggest_ideasread-only

    Starting points for this brand, each with a written prompt.

  • navigate

    Open another page of the app.

  • set_composer_mode

    Switch the composer between image, video, carousel, actor.

Making things

Every one of these writes into the composer on screen first, so the human watches the request being assembled, approves the spend on a card that shows the exact prompt, and then watches the result land on their wall.

  • draft_in_composer

    Write a prompt into the composer WITHOUT firing it - the human edits and presses Generate.

  • generate_imageneeds approval

    Render an on-brand image ad, product visual or social graphic.

  • generate_videoneeds approval

    Render a short video ad or lifestyle clip.

  • plan_carousel

    Plan a multi-slide deck and open the slide-by-slide review modal. Free until the human confirms.

Try saying

Things that work on the first try.

  • What is this site, and what would it cost me?
  • Read my brand and suggest five things worth making this week.
  • Draft an image prompt for a founder-holding-the-product shot, but do not generate it yet - I want to edit it.
  • Make two 9:16 image ads for my best-selling product, then tell me when they land.
  • Plan an Instagram carousel that explains why our returns policy is different.

Outside the browser

WebMCP is one of three ways an agent reaches Advibly, and they share the same backend. A hosted MCP server at advibly.com/mcp (OAuth 2.1, dynamic client registration) covers agents that are not in a browser, and npx advibly turns the same tools into terminal subcommands. Work started in one shows up in the others, because they are all writing to the same library.