Everything you need to run the agent

Connect an ad account, write a goal in plain language, and approve what the agent ships. The five-minute path from sign-up to your first optimization.

How to get reliable, safe, cost-effective results from an AI agent for ad ops — writing goals, guardrails, approvals, model choice, skills, schedules and true ROAS.

Concrete, multi-step jobs the agent runs end-to-end — audits, budget rebalancing, search-term mining, remarketing, PMax, and cross-account cleanups. Copy a goal and go.

How credits work, what consumes them, the free monthly allotment, top-up packs, where to see your usage, and how to lower your credit burn.

Connect your own tools to the agent over the Model Context Protocol (MCP), plus thousands of SaaS actions via Composio. Fields, steps, and how tools are discovered.

Give the agent 250+ SaaS app integrations — Slack, Gmail, Google Sheets, Salesforce, Notion and more — with one Composio API key. How to connect apps and have the agent act across them, with every write approved by you.

Connect an AI agent to Agent Planners over MCP, by hand: Claude (claude.ai, Desktop, Claude Code), Codex, Cursor, VS Code, Windsurf, Gemini CLI, ChatGPT, Grok, Muse and any MCP client — sign in from the agent or use an API key kept in an environment variable, verify the connection, add a second workspace, call tools and raw requests, and fix common errors.

Integrate your own software with Agent Planners: register an OAuth app (a server app with a client secret, or a public app with PKCE), send a person through the consent page once, keep each connection's tokens, and call the REST API — with the own-app rule (no approval step in its own workspace), raw access on by default, call logging, and publishing an app to other workspaces.

Call Agent Planners from your own system over plain HTTP: authenticate, discover what your credential can do, call the same platform tools as the MCP server, assemble a provider request where no tool exists, follow writes that wait for a person, read your own audit trail and credit usage — with an API key or an OAuth access token.

Instructions for Codex, Claude Code, Cursor and developers wiring another system to Agent Planners: choose MCP, an API key or an OAuth app; authenticate; discover with /api/v1/capabilities; call tools; handle 202 approvals and errors; verify — with a minimal Node and Python client.

How an OAuth app, an API key or an MCP agent calls Google Ads, Meta, TikTok, Microsoft, Reddit, Pinterest, Snapchat, LinkedIn and the other connected platforms in each platform's own request format — credentials injected, bound to the connected account, writes approved by a person.

The orchestrator plus six specialist agents, and how to choose the model for each one through the built-in AI Gateway — GPT, Claude, Gemini, Qwen, DeepSeek and more.

Point every agent at your own OpenAI-compatible endpoint — Azure AI, vLLM, OpenRouter or any self-hosted model. LLM tokens on your endpoint aren't billed as platform credits; you pay your provider directly.

Natively connect Google Ads, Microsoft Advertising, DV360, Campaign Manager 360, Search Ads 360, Meta Ads, TikTok Ads, Reddit Ads, Pinterest Ads, Snapchat Ads, LinkedIn Ads, X Ads, Google Ad Manager, AdSense, AdMob, Google Analytics, Search Console, BigQuery, Tag Manager, Merchant Center, YouTube, Klaviyo, HubSpot, GitHub, Shopify, SHOPLINE and Stripe — plus 250+ more apps (Gmail, Outlook, Slack, Notion, Salesforce…) via Composio and your own tools via custom MCP. Pick accounts with the context picker and run one task across an entire MCC.

Plan → approve → execute. How every write pauses at an approval gate, in-app or by email magic-link, with pre-approval, auto-approve, stale-state checks and the audit log.

Turn any goal into a recurring job on a cron — hourly, daily or weekly — in your timezone, re-evaluating against live data on every fire, with optional email approvals.

Alert rules email you when a metric crosses your line, when an ad account's status turns bad (suspended, verification required, disapproved ads, campaigns not serving), or when an AdSense site or AdMob app passes review.

Build a dashboard from a built-in template or have the agent design one, put it on a refresh schedule, and share it on a stable link you can revoke. How refreshes stay deterministic, what gets emailed, and what the share link does and doesn't do.

Agent Mail turns your inbox into a front door for the agent. Email a task from your sign-in address and it replies with what it would run, in which organization — nothing starts until you confirm. Covers the confirm flow, @mentions, standing rules, status checks and every limit that applies.

Capture your methodology as Skills, keep per-account memory (targets, protected campaigns, rules), and run isolated organizations with admin/editor/member roles.

Connect an International SHOPLINE store (your-store.myshopline.com) with two clicks — read store, product and order data, get an exact sales summary and anomaly detection, and make approval-gated store changes.

Connect a Shopify store in three tiers — domain-only storefront, read-only Admin data, and approval-gated Admin writes. How to create the Admin API access token, which scopes to grant, and the use cases the agent runs against a store.

Add Agent Planners to your home screen: it opens on your workspace, fits the notch and the keyboard, and shows an offline page instead of a browser error when the connection drops.

Turn an analysis or run into a client-ready PowerPoint deck and a written report — drafted by the agent from your real data, then editable and exportable.

A project groups the tasks, schedules, milestones and checklist items behind one goal, with its budget and KPIs, and an overseer agent that reads every channel's numbers and runs the optimization — moving budget, adjusting bids and creatives, sending reports and email. Every change follows the project's approval setting: by default each one waits for your approval, and auto-approve is a per-project opt-in.