Work from ChatGPT and Claude

Ask your AI to fix the account.You approve with a tap.

Claude, ChatGPT and other AI tools can read your accounts and propose fixes. Approval stays in Taseli.

Check my account for free

Free
Nothing changes without your approval

Why did cost per lead go up this week?

Google broadened matching, and 31% of spend moved to generic terms. I prepared 6 negatives.

Proposal: 6 negative keywordsAwaiting approval

What your AI can and can't do.

It can read and propose. It can never approve.

Can read	Can propose	Can never
Accounts, goals and currencies	Blocking searches	Approve anything, its own or anyone's
Results by campaign, date and segment	Budget changes	Skip your limits or the safety checks
Audit findings and open alerts	Targets, pauses and schedules	Write through queries (reading only)
Wasted searches and change history	Undoing a change	See payment details or other workspaces

Approval from inside a chat.

You ask
“Block the searches that wasted money last month.”
Your AI proposes
Taseli checks it and sends an approval link. Nothing changes yet.
You approve in Taseli
The link shows before and after. Your tap approves exactly that.
Your AI follows up
It tells you what ran, and that watching has started.

The tools

Every tool is clearly marked as read-only or a proposal.

All 17 tools

Tool	Type	What it does
`accounts.list`	Read	Accounts in the workspace with platform, currency and goals
`performance.get`	Read	Metrics by level, date range and segment
`gaql.run`	Read	Free-form read-only query (GAQL), up to 10,000 rows. Add-on.
`audit.latest`	Read	The latest audit findings
`alerts.list`	Read	Open alerts
`search_terms.wasted`	Read	Search terms that spent without results
`search_terms.ngrams`	Read	Which words inside search terms cost the most
`budget.pacing`	Read	Monthly pacing and forecast
`changes.history`	Read	Taseli's change journal and the platform's own changes
`recommendations.list`	Read	Taseli's and Google's recommendations, labeled
`proposals.create`	Proposal	Any change, validated and checked against policy
`proposals.negatives`	Proposal	Shortcut for blocking searches
`proposals.budget`	Proposal	Shortcut for budget changes
`proposals.status`	Read	A proposal's current state
`proposals.approval_link`	Read	The Taseli approval page - it can't approve
`proposals.rollback_request`	Proposal	Request a restore; needs your approval in Taseli
`playbooks.run`	Proposal	Run a playbook and get its proposals

Limits per AI, on top of yours.

Money limits
A daily cap on budget changes, in money and percent.
No flooding
A cap on proposals per hour, so WhatsApp stays quiet.
One change a day
The same campaign can't be changed twice in 24 hours.
Text is data, not orders
Ad text and searches are never treated as commands.

Connect in two minutes.

Claude, ChatGPT, Cursor, Gemini, and agents that run all day.

In Taseli, open Settings, then MCP, and copy your server address.
In your AI tool, add a custom connector with that address.
Sign in to Taseli and allow read and propose.
Ask your first question. Access lasts 30 days and can be revoked.

Plans

Reading comes with Pro. Proposing changes needs the Power add-on.

Pricing

Example: a client config

{
  "mcpServers": {
    "taseli": {
      "type": "http",
      "url": "https://<your-server-address-from-settings>"
    }
  }
}

Your exact server address is in Settings, then MCP.

Frequently asked questions

Why can't the agent approve changes?

Because the agent can be manipulated by the very content it reads - search terms, competitor ads, landing pages. Approval is a signed action in Taseli, bound to the exact changes, so a prompt injection can't spend your money.

Is this different from Google's own MCP server?

Google's official MCP is read-only. Taseli adds controlled writes: every change becomes a proposal with before and after, caps, a dry run, monitoring and restore.

Can an always-on agent run my account?

It can watch and propose around the clock. Anything that runs without your tap does so only under an autopilot policy you configured in Taseli - never because an agent asked.

What's logged?

Every call: the tool, the client name, the user and the result. You can see it in your audit log.

Does using MCP cost credits?

Reads don't - your own model does the work. Running a playbook does.

An AI campaigner. You just approve.

Taseli does the work. You make the call.