> ## Documentation Index > Fetch the complete documentation index at: https://docs.adsmcp.si/llms.txt > Use this file to discover all available pages before exploring further. # Autonomous Ad Ops Agents > Give ADSMCP a plain-English goal and a cadence. An autonomous Ad Ops agent monitors your accounts and proposes every change to an approval queue — nothing spends or changes without your sign-off. Autonomous **Ad Ops agents** turn ADSMCP from a tool you drive into a teammate that works on a schedule. You describe a goal in plain English, pick how often it should run, and the agent monitors your accounts, does the analysis, and **proposes every change to an approval queue**. You stay in control — nothing spends money or changes a live campaign until you approve it. > [!NOTE] > > Available on **Plus, Pro, and Max** plans. Agent runs count against your monthly tool-call quota, just like the tools they use. See [pricing](https://adsmcp.si/pricing). ## How it works **1. Describe the goal** In plain English: "Keep my Google Search CPA under \$40 and cut wasted spend," or "Watch all my accounts for broken conversion tracking." The builder compiles your goal into a concrete operating plan you can review. **2. Set guardrails** Choose the cadence (daily, weekly, or manual), a target metric, and hard limits: maximum budget change per run, maximum number of proposals, and any **protected campaigns** the agent must never touch. Widening scope or enabling money-moving actions requires an explicit high-risk confirmation. **3. The agent runs on schedule** On each run the agent pulls fresh performance data, analyzes it, and decides what to change — find winners to scale, search terms to add as negatives, budgets to rebalance, CPA/ROAS drift to correct, or broken tracking to flag. **4. You approve every change** Each proposed write lands in your **Approvals inbox** with the reasoning and the expected impact. Approve or reject from the agent card. Rejections take a one-tap reason chip that teaches the agent why — so it proposes better next time. Approved changes replay exactly once against your account. **5. The agent measures its own impact** After a change is approved, the **Impact** tab checks whether it actually helped, using your own account data — and feeds that verdict back into the next run. Verdicts only make a causal claim when your data supports it, so you won't see a misleading "+54% worse" call off a tiny sample. ## What an agent can do * **Find and scale winners** — identify high-ROAS campaigns and propose capped budget increases * **Cut wasted spend** — surface search terms with spend and no conversions, propose them as negative keywords * **Rebalance budgets** — shift spend toward what's working * **Guard CPA / ROAS** — catch drift against your target and propose corrections * **Flag broken conversion tracking** — so you're never optimizing blind * **Recommendations tab** — a running list of prioritized, ad-focused suggestions even before you turn on autonomy > [!NOTE] > > **The agent proposes; you dispose.** ADSMCP's autonomous agents never change your account or spend your budget without your approval. Every write is parked as a proposal until you approve it. This is a hard, server-side gate — not a setting you can accidentally disable. ## Cross-platform agents A single agent can reason and act across **Google, Meta, TikTok, LinkedIn, and Amazon** from one goal. For example: One cross-platform goal spanning three platforms. Create an Ad Ops agent that keeps my total monthly spend under 10,000 CAD across Meta, Google, and LinkedIn, rebalancing toward whichever platform is delivering the best cost per conversion. Run it daily and send every change to me for approval. ## Teach your agent Upload your own context — brand rules, target CPAs, seasonal calendars, standard operating procedures — and every run grounds its decisions in your documents instead of generic best practices. Your agent proposes changes the way *you* would. ## Community playbooks A **playbook** is an agent's decision logic — its goal brief, rules, and guardrails — with **no account data, budgets, or credentials attached**. You can: * **Share** a playbook you've tuned so other advertisers can use your approach * **Fork** someone else's proven playbook onto your own account in one step Only the strategy travels; your accounts and spend never do. Install counts show which playbooks the community trusts. > [!NOTE] > > Community playbooks (shareable agent strategies) are different from [Community Plugins](https://docs.adsmcp.si/agent-skills/community-plugins.md), which are open-source analysis *tools* that extend ADSMCP's capabilities. ## Notifications & approvals Proposals appear in your in-app **Approvals inbox**, and the agent can email you when it has something waiting. Every notification states the real change and, if a run failed, the real reason — grounded in what actually happened, not a canned template. Money-moving actions always route back to an explicit in-product approval. ## FAQ > **Can an agent spend my budget without asking?** > > No. Every write — budget change, pause, new keyword, new campaign — is parked as a proposal and only executes after you approve it in the Approvals inbox. This is enforced on the server, not just in the UI. > **What's the difference between an Ad Ops agent and Agent Skills?** > > [Agent Skills](https://docs.adsmcp.si/agent-skills/overview.md) teach your AI client the right *workflow* when you're actively chatting with it. An **Ad Ops agent** runs on its own schedule in the background, without you in the chat, and queues its proposals for later approval. Skills make your live sessions expert; agents keep working when you're away. > **Which plans include Ad Ops agents?** > > Plus, Pro, and Max. Agent runs use your monthly tool-call quota. Higher tiers get more quota for more frequent runs across more accounts. > **Can I run one agent across multiple ad platforms?** > > Yes. A single agent can reason and propose changes across Google, Meta, TikTok, LinkedIn, and Amazon from one goal — for example, capping total spend across three platforms at once. ## Related > **[Community Plugins](https://docs.adsmcp.si/agent-skills/community-plugins.md)** > > Extend ADSMCP with open-source, read-only community analysis tools. > **[Agent Skills Overview](https://docs.adsmcp.si/agent-skills/overview.md)** > > Teach your AI client proven ad-management workflows. > **[Core Workflows](https://docs.adsmcp.si/agent-skills/workflows.md)** > > Step-by-step sequences for campaign creation and optimization. > **[Pricing & Plans](https://adsmcp.si/pricing)** > > Quota by tier for agent runs. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.