- Library
- Edition 2.0
- Claude Ads
Edition 2.0
Connecting Claude to Meta ads, route by route.
The supported ways to connect Claude to Meta advertising in 2026, from Meta’s hosted MCP server to your own Marketing API integration, with the exact credential to use in development and in production.
For technical media buyers, agencies and automation developers. About 170 minutes across 12 chapters. Facts checked .
Replaces the Claude Ads Safety Guide (V1). Links to the first edition now open this guide.
Get the full guide
Free. 12 chapters, about 170 minutes. One sign-up opens all three Edition 2.0 guides on this device.
Chapters
12 chapters. Take them in order; each one ends with a task and a checklist.
- 01
Choose your route
The three supported ways to connect Claude to Meta ads, the five patterns to stop using, a ten-minute vetting checklist for any tool, and how to pick a starting route and credential for your own ads or your clients' ads.
12 minDone - 02
Set up your business portfolio, app and sandbox
The Meta groundwork every route needs, click by click. Check the business portfolio, start business verification if you will serve clients, create a Meta app with the right use case, and create a sandbox ad account where test writes cannot spend.
14 minDone - 03
Route A: connect Meta's ads MCP server
Connect Claude apps and Claude Code to Meta's hosted ads MCP server, set each tool's permission, block activation and audience uploads, add Meta's server rules as a cap Claude cannot talk around, and check your first paused draft in Ads Manager.
15 minDone - 04
Route B: a development setup that cannot spend
Run Marketing API calls and Ads CLI commands from your own terminal against a sandbox ad account. Get Explorer tokens, exchange a longer one when you must, inject secrets per command, and lock Claude Code down so it writes the code but never holds a Meta token.
14 minDone - 05
System users and tokens, click by click
Create system users that each do one job, assign only the assets they need, issue 60-day expiring tokens straight into a secret manager, and refresh, rotate, revoke and inspect those tokens with Meta's documented calls, without Claude ever seeing one.
14 minDone - 06
Route C: production access and limits
What Meta requires before an integration touches live advertisers. Access levels and the Full tier, App Review screencasts that match Meta's stated requirements, publishing, Require App Secret, the rate limits and per-object limits that shape your design, versions and SDKs, and your policy obligations.
13 minDone - 07
Marketing API essentials and the debugging playbook
The details that decide whether an API write is safe. Objectives, the required special ad category, paused status, money units, Advantage+ changes, attribution and timezone, writes that are not idempotent, and a playbook that maps each Meta error to its first check.
14 minDone - 08
Agencies and multi-client access
Take on client ad accounts safely. Clients grant access through partner access or a login configuration, your services bind every request to one client, credentials and AI context stay separate per client, portfolio views use flags not pooled data, and onboarding, freezing and offboarding follow checklists.
15 minDone - 09
Build the read-only broker
The read side of Route C. Narrow, client-bound, read-only tools that return aggregates with their data contract, a tool loop that never hands Claude a credential, what the MCP specification requires if you expose the broker as a connector, and the attack tests it must pass.
14 minDone - 10
Proposals, approval and the executor
The write side of Route C. A proposal schema Claude must fill, an approval card that shows every row, an executor that refuses by default and re-reads Meta before each write, a journal that makes each write run once, Meta-specific caps, pre-activation checks and four release gates.
15 minDone - 11
Monitoring, audits and incidents
Run a Claude and Meta integration month after month. Monitor the controls as well as the campaigns, keep an operations calendar, run the 30/60/90 audit cycle, respond to a leaked credential in the right order, and handle a Meta restriction through the official path.
15 minDone - 12
Prompt library and the safe-drafting skill
Every V1 prompt rewritten for today's tools, grouped by job. Learn an account, report, diagnose, draft paused objects, propose changes, debug and research competitors. Each prompt states its inputs and what a good answer looks like, and every write goes through approval. Plus one skill and the anti-patterns to delete.
15 minDone
What changed since V1
- Meta now hosts an official ads MCP server (open beta since 29 April 2026) that can create, edit and activate ads. The confirmation step lives in your AI client.
- Development and production use different Meta credentials on purpose: short-lived tokens on the Limited tier for testing, 60-day expiring system users on the Full tier for production.
- Claude’s connector directory now serves Claude apps, Claude Code and Cowork, and Anthropic is explicit that verification is not a security audit.
- The MCP authorization spec requires tokens bound to the server they are for and forbids passing them through to other services.
Review status
Beta Available for testing; behavior and interface may change. Next review due .
- Owner review of every chapter before the status moves from beta to active
- Re-verify connector availability and permissions in the Claude directory before each recording
- Re-verify Meta token and access-tier facts at each Marketing API release
Primary sources used in this guide
Official documentation and changelogs only. When a source and this guide disagree, the source wins.
- Meta ads AI connectors announcement Meta
- Manage ads from an AI agent with Meta ads AI connectors Meta
- Ads MCP server overview Meta
- Ads CLI, get started (system user setup) Meta
- Marketing API authorization and access tiers Meta
- Business Verification for apps Meta
- How to verify your business in Meta Business Suite Meta
- Meta Terms of Service Meta
- Automated Data Collection Terms Meta
- Meta Platform Terms and Developer Policies Meta
- Connectors directory Anthropic
- Add a connector that isn't in the directory Anthropic
- Claude Code skills Anthropic
- Using Claude in Chrome safely Anthropic
- npm scripts and lifecycle events npm
- Give a partner access to business assets in your business portfolio Meta
- Create an app Meta
- Marketing API use cases (publishing) Meta
- Marketing API sandbox capability re-enabled Meta
- Ads MCP server, get started Meta
- Securing requests (app secret location) Meta
- Marketing API rate limiting (per-object limits) Meta
- Ads MCP server ad creation and management tools Meta
- Ads MCP server activity log tools Meta
- Ads MCP server rules best practices Meta
- Get started with connectors Anthropic
- Authentication for connectors Anthropic
- Claude Code MCP (requiresUserInteraction and organization controls) Anthropic
- Claude Code permissions Anthropic
- Meta Business Tools Terms Meta
- Graph API Explorer guide Meta
- Access tokens Meta
- Long-lived access tokens Meta
- Ads CLI overview (effective_status) Meta
- Claude Code sandboxing Anthropic
- Claude Code security Anthropic
- Claude Developer Platform workspaces Anthropic
- 1Password CLI op run 1Password
- Add system users to your business portfolio Meta
- About system users in Meta Business Suite Meta
- System users Meta
- Install apps, generate, refresh and revoke tokens Meta
- debug_token reference Meta
- Graph API error handling Meta
- Marketing API changelog (access tier rename) Meta
- Permissions reference and App Review Meta
- Marketing API versioning Meta
- Marketing API versions Meta
- Meta Business SDK for Node.js releases Meta
- Campaign reference Meta
- Ad set reference Meta
- Currencies Meta
- Ad account insights reference Meta
- Advantage+ campaigns Meta
- Marketing API v25.0 changelog Meta
- Marketing API v26.0 changelog Meta
- Out-of-cycle changes 2025 (attribution windows, data limits) Meta
- Out-of-cycle changes 2026 (breakdowns) Meta
- Remove partners from your business portfolio Meta
- Facebook Login for Business Meta
- MCP authorization specification 2025-11-25 Model Context Protocol
- Tool use overview and strict tools Anthropic
- Models overview Anthropic
- Structured outputs Anthropic
- Troubleshoot a disabled or restricted account Meta
- About Meta Business Support Home Meta
- Conversions API deduplication Meta
- Generative AI transparency in Meta's ads products Meta
- How Meta is preparing for the 2026 US midterms Meta
- Anthropic Usage Policy Anthropic
Education, not legal, security or financial advice. Never paste a live access token, API key or customer data into an AI chat, a prompt, a screenshot or a public repository.