Bring your own AI

GoodLads, inside the AI you already use.

One URL, one approval, and Claude, ChatGPT, Cursor or anything else that speaks MCP can read your Google Ads accounts - and reason about them the way GoodLads does, because the playbooks come with the data. What it lands on, it can put on your board as a proposal; deploying it is still your click here. Your assistant does the thinking on your tokens, so there is no usage limit and no extra bill from us.

https://goodlads.cc/api/mcp

Connect it

You need a GoodLads account with at least one Google Ads account connected. Everything else is the client asking you to log in once.

Claude (web, desktop, mobile)

Paste one URL in Settings, approve once, done.

Open Claude connectors

Settings - Connectors - Add custom connector.
Paste this URL:
```
https://goodlads.cc/api/mcp
```
Claude sends you to GoodLads to log in and approve access. Approve, and the tools are there in every chat.

Claude Code

One command.

Run it in any project:

claude mcp add --transport http goodlads https://goodlads.cc/api/mcp

Then /mcp in Claude Code to run the OAuth login.

Cursor

One click, then approve.

Add to Cursor

The link opens Cursor's install prompt with the endpoint filled in.
Or paste it yourself under Settings - MCP:
```
https://goodlads.cc/api/mcp
```

VS Code

One click, then approve.

Add to VS Code

The link opens VS Code's MCP install prompt with the endpoint filled in.

ChatGPT

Same URL, in the connector dialog.

Settings - Connectors - Advanced - Developer mode, then Create.
Paste this URL and log in when it asks:
```
https://goodlads.cc/api/mcp
```
ChatGPT does not show MCP prompts, so use the tools directly: ask it to call list_playbooks first.

Anything else that speaks MCP

Streamable HTTP, OAuth 2.1, no API key to manage.

The endpoint:
```
https://goodlads.cc/api/mcp
```

Or, for a client that reads a config file:

{
  "mcpServers": {
    "goodlads": {
      "type": "http",
      "url": "https://goodlads.cc/api/mcp"
    }
  }
}

It brings the method, not just the numbers

An assistant with raw Google Ads rows will give you plausible advice about a live budget, which is the expensive kind of wrong. So the server ships the same playbooks GoodLads holds its own generated ideas to - one per lever, each naming the evidence to pull, the bar the idea has to clear, and the honest mechanics of the change. Your AI calls list_playbooks, then get_playbook, before it looks at a single figure.

Limited by budget

The campaign is capped - test a lower target CPA so the same budget buys more conversions.

Applies whenSearch-family campaign (an experiment can run) with a budget snapshot to reason about. App campaigns get app_bid_target instead.

Ad-group bidding targets

Tune one ad group's target so the bidder buys more of what converts and less of what does not.

Applies whenSearch-family campaign (an experiment can run), more than one ad group, and a bidding strategy carrying a tunable target (Target ROAS or Target CPA). Without a target there is no knob to turn.

App campaign targets

App campaigns cannot run experiments - step the live target directly, and put it back in one click.

Applies whenApp campaign with a campaign-level Target CPA or Target ROAS that is NOT a shared portfolio strategy - a portfolio target belongs to other campaigns too, so the write path refuses it.

Geography

Some places convert far better than others - stop paying the same for the ones that do not.

Applies whenSearch-family campaign (an experiment can run) that actually served at least two places, so there is a spread to act on rather than a single-city average.

Keyword-ad combinations

Pair proven keywords with ads written for them - tighter keyword-to-ad relevance.

Applies whenA SEARCH campaign with at least one search term of its own, or a routing finding from the account-wide analysis. With neither there is no proven keyword to build an ad around.

Messaging angle tests

Test one untried angle against the ads running today, and get a graded read on it.

Applies whenA SEARCH campaign with at least one live responsive search ad, so the incumbent angle can be named from real headlines rather than guessed at. The channel half is not decoration: the ad-copy read returns App ad text too, so without it this skill runs on App campaigns and writes a responsive search ad they cannot take.

Performance Max asset experiments

Test a new argument against the one this campaign already makes - same asset group, same images, same audience, split traffic and a real read.

Applies whenA Performance Max campaign with at least one enabled asset group whose headlines were written by the advertiser (so there is copy to run as the control arm) and which has room under Google's cap of 15 headlines for the challenger lines. Retail Performance Max is included: an asset experiment adds no asset group, so there is no product filter to overlap.

Ad-group splits

Split an ad group that mixes search intents, so each theme gets an ad written for it.

Applies whenSearch-family campaign, and the split detector nominated at least one ad group as mixing several search intents. No nominated ad group means there is nothing to split.

New campaigns

Proven demand this account cannot serve well from the structure it has today.

Applies whenA completed research report, a landing page to point at (from the campaign or the report), and at least one search term already proven by real conversions and real spend.

Negative keywords

Cut spend on search terms that never convert, so the budget flows to terms that do.

Applies whenAlways runs when the account has this skill enabled. Account-scope: it works the whole roster at once, so there is no per-campaign precondition to check.

Ready-made passes

In clients that show MCP prompts (Claude, Claude Code and Cursor; ChatGPT does not), GoodLads adds them as commands you pick rather than sentences you have to write:

GoodLads account review - The full pass GoodLads makes over an account: every lever, in order, with the evidence each one needs and nothing padded.
Where is this account leaking money - The waste pass: search terms taking real money and returning nothing, and terms competing against themselves in two ad groups.
Pressure-test an idea before spending on it - Take an idea - yours, an agency's, or one on the GoodLads board - and try to break it against the account's own numbers.
One per lever - 10 more, so you can run a single playbook against a single account without explaining what you want.

Or ask in your own words. The five jobs worth starting with:

Using your own AI has the exact prompt for each, and what your assistant does behind it.

What it will not do

Touch your account. There is no tool on the server that can change a campaign, an ad, a budget or a keyword, whatever your assistant is asked to do, and a test walks the code's import graph on every build to keep it that way. It can write one thing: a proposal on your GoodLads board, the same card the product writes for itself, with the draft filled in and the evidence attached. Deploying that card is yours alone: you open GoodLads and press the button, with the usual one-click approval and a recorded way back.

Access is your own login, through OAuth: you approve it on a GoodLads consent screen, the client holds the token, and it sees exactly the accounts you have connected and nothing else. Revoke it by removing the connector in your client.

It also costs you nothing extra. Your assistant does the reasoning and pays for its own tokens, so there is no usage limit on the endpoint and no second bill from us.