Agent PlannersAgent Planners
Integrations

Connect Agent Planners to OpenAI Codex: ad-platform tools in your terminal

TL;DR

Codex reads MCP servers from ~/.codex/config.toml, and Agent Planners is a remote MCP server over Streamable HTTP, so one command adds it and one more signs in. From then on Codex can read your ad and analytics accounts, or hand the platform a whole goal and poll it, without ever holding a platform credential.

How to connect

Add the server to Codex, then sign in:

codex mcp add agentplanners --url https://www.agentplanners.com/api/mcp

codex mcp login agentplanners

The login opens the Agent Planners consent page in your browser: sign in, choose the workspace and approve what Codex may do; Codex stores and refreshes the tokens. You can also write the server into ~/.codex/config.toml under [mcp_servers.agentplanners] with the same url, or use an API key instead of signing in by naming an environment variable that holds it in bearer_token_env_var. Check codex mcp list to confirm it registered — a typo in the [mcp_servers.name] header makes Codex skip the section without an error, so verify rather than assume. OAuth sign-in has the details.

What it is useful for

  • Pulling real campaign or GA4 numbers into whatever you are building, without leaving the terminal.
  • Checking an account before a release — spend, conversions, anomalies.
  • Handing a long job to the platform's own agent and polling it while you keep working.

Frequently asked questions

Does Codex need any ad-platform credentials?
No. The platform holds the connections; Codex authenticates only to Agent Planners, with its OAuth sign-in or an API key.
Is there a Codex directory listing?
Not yet. Codex installs plugins from marketplaces — OpenAI's curated one and any you add with codex plugin marketplace add — and Agent Planners is not in OpenAI's curated list. The two commands above connect it today.
Can I give a teammate read-only access?
Yes — the person connecting can untick everything but the read permissions on the consent page, or you issue them a key with read scopes only. Scopes are per connection.
Connect Connect Agent Planners to OpenAI Codex and try it on your own account

Start free — 2,500 credits a month, no credit card. Reads are free; every write waits for you.

Related platforms

Agent Planners MCP server
A remote MCP server (Streamable HTTP) that lets Claude, Codex or any MCP host run tasks and call ad-platform tools under your connected accounts — with the same scopes, guardrails, approvals and audit trail as the app.
Connect Agent Planners to Claude
Add Agent Planners as a custom MCP connector in Claude Desktop or Claude Code and ask Claude to audit campaigns, mine negatives or pull GA4 reports — under your own connected accounts, with writes governed by the key's write mode and your organization's guardrails.
AI agent for Google Ads
Connect Google Ads (including MCC) and let an AI agent audit performance, mine negatives, rebalance budgets and build campaigns — every write pauses for your approval.

How the agent works

Getting started
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.
Approvals & workflow
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.
Agent playbooks
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.
Connect your platforms
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, GA4, 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.