Write access in MCP

What it is

Most early MCP servers were read-only: they pulled reports, summarised and searched. Every chat ended with the same sentence: “now go into the dashboard and do this.” Write tools close that loop; the model proposes the change and applies it through the platform's official API.

MCP describes this with tool annotations: readOnlyHint, destructiveHint, idempotentHint and openWorldHint. These are the server's declaration, not a technical constraint; they state what a tool claims to do, they do not prevent it. Your trust in the server remains decisive.

Why it matters

Write access puts analysis and action in one place, but mistakes on a live budget are expensive: the right change in the wrong account, a budget read as 400 instead of 4,000, a pause applied too broadly.

Responsible write access has a recognisable shape: OAuth scopes narrowed to least privilege, per-customer isolation, correct account scoping, a preview and approval before execution, a cap on how large a single change can be, and an audit log. A mature server returns API errors readably and never leaves a half-applied change.

How Tecrube uses it

This is why Tecrube exists: creating and pausing campaigns, budgets and bids, keywords and negatives, audiences, creatives, extensions and conversion actions are live on Google Ads, Meta Ads, Microsoft Advertising, TikTok Ads and LinkedIn Ads.

Every write action returns a preview first, is applied only on your explicit approval, rejects any single budget increase above 50%, and is written to the audit log. Per-customer isolation and MCC support are standard on every plan.

Example

“Find the ad groups in my non-brand campaign whose CPA is 40% above target and cut their bids by 15%.” The assistant lists the ad groups and previews the current and proposed bid for each; when you say “approve”, it applies and logs them.