Bolrach MCP

The Bolrach MCP lets an AI assistant run a Bolrach Ads account for its owner. Someone might ask Claude for a spring sale campaign, or ask ChatGPT which of last week's two headlines did better. Both requests end up at the same API our own console calls, so the assistant sees what the owner would see.

Address https://mcp.bolrach.com/mcpTransport Streamable HTTPSign-in Authvio (OAuth 2.1)

The address

https://mcp.bolrach.com/mcp

Nothing else needs typing. From that URL alone, the client finds Authvio, registers itself and opens the sign-in page.

Connecting a client

ClientWhere
ClaudeSettings, Connectors, Add custom connector, then paste the address
Claude CodeThe claude mcp add command below, then /mcp to sign in
ChatGPTDeveloper mode on, then Settings, Apps and connectors, Create, with OAuth
OthersAny client with Streamable HTTP and OAuth support
claude mcp add --transport http bolrach https://mcp.bolrach.com/mcp

Each one ends on the Authvio consent screen. The person signs in, looks over the permissions and presses Allow. With Claude Code the browser opens on the developer's own machine and the token never leaves it.

The tools

GroupTools
Accountslist_ad_accounts, create_sandbox_account
Campaignslist, get, create, update, set budget, activate, pause, resume, delete
Ad groupslist, create, update, set targeting
Creativeslist, create, update, delete, upload image, draft ad copy
A/B testsstart, read results, auto-pause on or off, stop
Reportsget_campaign_report

Look closely at a tool in Claude and a small label says whether it only reads or will add or change something. Claude and ChatGPT both put that flag in front of the person before running it. Deleting and spending get a confirmation prompt in clients that offer one.

Permissions on the consent screen

PermissionWhat the assistant may do
ReadSee accounts and campaigns
ReportsRead results and A/B tests
WriteCreate and change drafts, run sandbox tests
PublishStart or resume a campaign, test on a live account, raise a running budget

Publish is the one that spends. Many people leave it off at first. The assistant then gets a campaign ready, and the owner presses start in the console after looking it over.

An assistant reaches the accounts its person can reach at bolrach.com/ads, with the same rights and no more. Each change lands in the account's audit log under that person's name and the client's. The Ads API page lists every permission in detail.

A safe first run

The first request worth making is "make me a sandbox". Campaigns there see test placements only and cannot spend a cent. Watching an A/B test take shape in one tells a new user more than this page does.

Disconnecting

Open the Authvio account, go to Connected apps and remove the client. The token dies with it.

What the server does not do

The server shows no ads of its own and trains no model on anyone's account. It stores no campaigns either; each answer is read fresh from Bolrach Ads.

Every endpoint with its parameters: the Bolrach API reference. Machine-readable: https://api.bolrach.com/v1/openapi.json.