Start free
    Menu
    How it worksIntegrationsCustomersPricingSign in

    how it works

    from raw data
    to a shipped change.

    Cogny reads your marketing data, proposes one next step, makes the change once you approve it, and comes back to measure it. Then a verdict. Then the next one.

    Start freeSee pricing
    claude mcp add --transport http cogny https://app.cogny.com/mcp

    One line in Claude Code, then authorise in the browser. Also ChatGPT, Cursor and Codex, or the Cogny app with no terminal.

    example ticket
    1. you askWhy are signups flat this month?
    2. it finds/pricing sits 11th on Google. The title never says what it is.
    3. you approveRewrite the title in WordPress.
    4. it shipsLive on Oct 7. Measured Oct 14.
    5. partialclicks +18%, signups flat

    one loop.

    The whole path in one place: no analyst queue, no week-long turnaround. Minutes from a question to a change, then days to an answer.

    1. 01connectLink your ads, analytics, search, email, CRM and site. OAuth or an API key, from the app or from your own AI.
    2. 02askAsk in plain words. Cogny reads the raw data across your tools and answers with one concrete next step, not a deck.
    3. 03approveYou say yes: in the Cogny app, or in the approval prompt of your own AI.
    4. 04shipThe change is made in the tool itself: the budget, the keyword, the post, the page, the tag. It becomes a ticket with what changed and what should happen.
    5. 05measureThree days later for ads, seven for content and SEO, Cogny compares before and after in your own connected data, wherever the result shows up.
    6. 06verdictThe ticket closes as worked, partial, didn't work or inconclusive. Then the next step.
    7. ↻againWhat worked becomes the next question. A finished chat can become a loop that runs on a schedule.

    every change gets a verdict.

    Misses included. On every plan, Free too. A measured win gets a share card.

    • workedMoved the number it was meant to move.
    • partialSome of it did.
    • didn't workKept on the record, not buried.
    • inconclusiveToo little data. Checked again, then closed after four tries.

    How does Cogny measure a change?

    When a change ships, Cogny opens a ticket and reads the result in your own data 3 days later for ads and 7 days later for content and SEO. It closes the ticket as worked, partial, didn't work or inconclusive. This runs on every plan, Free included, and even when a workspace has no credits left.

    Does Cogny change anything without asking?

    No. In the Cogny app, write tools ask first by default, and you set that per tool. From your own AI over MCP, that AI's approval prompt is the gate; Cogny doesn't ask a second time. A tool set to blocked is refused wherever the call comes from, and every write is logged with the tool, the time, the result and who authorised it.

    Can I run Cogny from Claude, ChatGPT, Cursor or Codex?

    Yes. Add Cogny as an MCP server: in Claude Code, run the one install line and authorise in the browser; Claude.ai, ChatGPT, Cursor and Codex connect the same way. One connection reaches every integration you connected. Launches and budget, bid and status changes your AI makes become measured tickets on their own; for other changes it offers to register them.

    across your whole stack.

    A change in one tool, measured where the result shows up, often in another. 72 integrations, 1,375 read and write tools.

    1. Pause broad-match keywords with no conversionsGoogle Ads measured in Google Ads3 daysworked
    2. Rewrite the pricing page title and introWebflow measured in Search Console7 dayspartial
    3. Move retargeting budget to videoMeta Ads measured in Google Analytics 43 daysdidn't work
    4. example

    Google Ads · Meta Ads · LinkedIn Ads · TikTok Ads · Google Analytics 4 · Google Search Console · Google Tag Manager · Shopify · Klaviyo · Mailchimp · WordPress · Webflow · every integration and tool

    run it from your own AI.

    Or from the Cogny app. Same tools, same loop. Cogny keeps the ledger.

    your own ai

    Claude, ChatGPT, Cursor or Codex.

    One MCP connection reaches every integration you connected. Your AI does the thinking and asks before it changes anything. Cogny logs every write, and the campaigns and ads your AI launches become measured tickets on their own.

    Connect your AI

    the cogny app

    Chat or talk. No terminal.

    Cogny's own agent proposes the step, makes the change when you approve it, and opens the ticket that gets measured. It starts on a graph of your numbers against the previous 30 days.

    Open the app

    On every plan, launches your own AI makes are tracked automatically. For other changes, it offers to register them, and they are measured the same way. On Cloud, your AI can also work an assigned ticket from start to finish.

    you approve. it's logged.

    • In the Cogny app, write tools ask first by default. You set it per tool.
    • From your own AI, its approval prompt is the gate. Cogny doesn't ask a second time.
    • Set a tool to blocked and it is refused, wherever the call comes from.
    • Every write is logged: the tool, the time, the result and who authorised it.
    • On Cloud, owners choose which integrations and tools each member can use.
    • On Cloud, code changes arrive as pull requests. A person merges them.

    the good ones run again.

    solo · cogny ai · cogny pro

    Loops.

    Turn a finished chat into a loop that runs on a schedule, with the tools you pick, from the app or from your own AI. Each run is its own ticket, measured like the first. Up to 5 per workspace.

    cloud

    The team version.

    Scheduled reports, tickets generated from them across every channel, up to three proposed experiments for each graph, and Cogny Code for changes to your site as pull requests.

    start with one change.

    Start freeCloud for teams · book a demo

    No card for Free. Compare plans

    Agency? A workspace per client →