MCP Server
One endpoint, added once, and the assistant you already use can answer from your live ad data — spend, search terms, auction insights, raw GAQL. 57 tools sit behind it. 45 of them are reads, granted the moment you connect. Changes are a separate scope your client asks you to approve.
Server URL
https://adscriptly.com/api/mcpThe same URL, the same OAuth handshake, the same tools. Pick the client you already work in — the steps differ, the server does not.
Claude
Ask about spend, search terms and auction insights mid-conversation.
InstallClaude Code
Your ad accounts in the same terminal as your codebase.
InstallChatGPT
Reach the same tools from a ChatGPT conversation.
InstallOpen Claw
Long-running agents that keep working after you close the tab.
InstallGrok
Same server, same scopes, inside Grok.
InstallHermes
Self-hosted agents, connected over the same OAuth handshake.
InstallInstall
Pick your client. Where the vendor publishes a one-click install link, that is the button. Where they do not, one press copies the exact string and opens the screen you paste it into — never a page of setup.
Server URL
https://adscriptly.com/api/mcpClaude
One click opens Claude with the connector dialog already filled in.
Signed out? Claude signs you in first and returns you to the filled-in dialog. Team and Enterprise admins add it once for everyone from Organization settings → Connectors.
Select Add to Claude. The Add custom connector dialog opens with the name and URL already filled in.
Review the values and select Add.
Sign in to Adscriptly when the OAuth window opens.
Or add it by hand
https://adscriptly.com/api/mcpEvery client that reads MCP servers from a JSON file takes this block. The server name is adscriptly; the transport is Streamable HTTP.
{
"mcpServers": {
"adscriptly": {
"type": "http",
"url": "https://adscriptly.com/api/mcp"
}
}
}Also one click, in your editor
What happens next
Adding the server needs no plan. New here? You create a free Adscriptly account and there is a short profile step on the way through, then you land back in the connector with nothing to redo. Either way the connector comes back healthy and your client lists every tool it can run.
The first time you actually ask it something, the server replies with an activation link instead of the data, because that answer is read out of your live ad account and that is the part a plan pays for. It arrives as an ordinary answer, not an error — a sentence and a link your assistant reads out to you, rather than the red failure an expired key would produce.
Start the trial and ask again. The connection stays exactly as it is — nothing to reinstall, nothing to reconnect.
Free for 7 days, then $129/mo. We'll remind you 3 days before your first charge. Cancel anytime. Card required.
Every name below is a tool the server actually exposes. The model picks the one your question needs; you never learn the names.
Ask anything
Raw GAQL, resource discovery, and queued report generation with bounded status checks.
Where the money went
Spend and outcomes at every level, without exporting a single CSV.
Who saw it
Every segment Google will break performance down by.
Creative and assets
Down to which RSA headline is carrying the ad group.
Competitive and diagnostic
Impression share, overlap, and who changed what, when.
Conversions
Including the outcomes Adscriptly sent back, not just the ones Google counted.
Changes, behind approvalapproval
A separate scope, annotated destructive, surfaced by the client for approval.
Read and write are different scopes, and the client is told which is which.
An account, yes. A plan, no — and that holds whether or not you have used Adscriptly before. With an existing account, connecting is the sign-in you already have and nothing else. If you are new, you create the account during that same sign-in — there is a short profile step — and then connect, still without picking a plan. Either way the handshake completes and the client lists all 57 tools. Entitlement is checked only when a tool actually runs, so the first query is where a free trial or a plan is needed, and the tool hands back a link to start one instead of an error.
Any client that speaks remote MCP over HTTP. Claude, Cursor and VS Code publish an install link, so one click opens their connector dialog with the name and URL already filled in — you still review and confirm it, and nothing installs silently. Claude Code, ChatGPT, Open Claw, Grok and Hermes take one copy and a paste, because their vendors publish no such link. Anything else that accepts a server URL connects the same way: there is one endpoint and it is the same for every client.
OAuth 2.1 with Dynamic Client Registration and S256 PKCE. The client registers itself, you sign in to Adscriptly once from inside it, and no API key is ever pasted into a config file. Your Google credentials are not handed to the model: the server holds the connection, the client holds a token for Adscriptly.
Not by default. Reads and writes are separate scopes, and a connected client is granted the read scope — that is 45 of the 57 tools. Writes run through execute_mutate, which is annotated as a destructive operation, so the client surfaces it for approval before anything changes in the account.
The tool catalogue is Google Ads. For Meta, Adscriptly delivers conversions through the Conversions API and reads accessible accounts with the ads_read permission — it does not request campaign-management permission, so Meta campaigns cannot be edited from a chat. /claude-meta-ads covers exactly what is and is not available there.
There is no separate charge for the MCP server — it is included with any paid plan and with an unexpired free trial. Every tool call checks entitlement, so an expired trial stops answering and hands back a reactivation link rather than degrading quietly.
Free for 7 days, then $129/mo. We'll remind you 3 days before your first charge. Cancel anytime.