Meta Ads Integration with ChatGPT Using MCP: Complete Step-by-Step Guide

Updated: October 31, 2026
Audience: Beginners, advertisers, business owners, agencies, marketing analysts, and developers
Level: Beginner to intermediate
Integration: Meta’s official hosted Ads MCP server (https://mcp.facebook.com/ads)

SEO publishing information

Verification note (October 31, 2026): Meta Ads MCP is in beta. The Plugins connection flow supersedes older Apps / Developer Mode steps. Meta developer pages may require login, so the full current tool list and a live account connection could not be independently verified. This guide distinguishes official documentation from third-party reporting and does not claim a live write test.

1. Introduction: What are we connecting?

Meta Ads is the advertising system behind campaigns shown on Facebook, Instagram, and other eligible Meta placements. Advertisers usually work in Ads Manager, selecting a campaign objective, defining an audience, setting a budget, uploading creatives, and reviewing delivery and business outcomes.

ChatGPT is an AI assistant that can summarize data, explain trends, compare options, and, when an appropriate external tool is connected, request actions in other services. ChatGPT cannot read your private advertising account merely because you mention its name in a conversation: it needs an authenticated integration with sufficient permissions.

Model Context Protocol (MCP) is a standard interface through which an AI client discovers and calls tools provided by an external server. In this case, Meta hosts the remote Ads MCP server at:

https://mcp.facebook.com/ads
Code language: JavaScript (javascript)

You provide this URL to a compatible ChatGPT custom MCP connection, authenticate with Meta, and then ask questions about accounts you are authorized to access. Meta introduced Ads AI Connectors in open beta in April 2026; its own newsroom subsequently described using third-party AI tools to create, manage, and analyze campaigns. Meta newsroom

Ads Manager versus ChatGPT with MCP

The advantage is faster analysis, not better advertising. Check attribution and approve all spending changes.

2. Prerequisites and eligibility

Checklist before starting:

• [ ] A ChatGPT account and access to ChatGPT on the web.
• [ ] A plan and workspace role that permit creating or using custom MCP server connections.
• [ ] A Facebook account or Meta Managed Account that can sign in to the business.
• [ ] A Meta Business Portfolio with at least one advertising account.
• [ ] Appropriate Ads Manager permissions for each ad account you want to inspect.
• [ ] Permission to authorize a third-party AI connection under your company’s policy.
• [ ] A current browser, working internet access, and access to the login or two-factor challenge if Meta requires it.
• [ ] Agreement on whether the connection should be read-only or allowed to change campaigns.
• [ ] A small, non-critical account or paused campaign to use for initial testing.

Which ChatGPT plans support MCP?

OpenAI documents full custom MCP read/write access for Business and Enterprise/Edu, subject to workspace policy. Business admins/owners create and publish connections; Enterprise/Edu admins can authorize selected developers through RBAC. Pro has documented read/fetch-only custom MCP access. Custom creation is not established here for Free, Go, or Plus. Confirm your account controls. OpenAI MCP eligibility

Meta side: Standard hosted authorization generally needs no developer app. Custom OAuth clients or applications serving other businesses may need a Meta app and review; consent prompts vary. Meta getting started

3. How the integration works

ChatGPT interprets your request, MCP transports a tool call, and Meta verifies authorization before returning data or an action result.

flowchart TD
    U[Advertiser types a request] --> C[ChatGPT and MCP client]
    C -->|Tool discovery / call over HTTPS| M[Meta Ads MCP server]
    C <-->|User sign-in and consent| O[Meta OAuth / Facebook Login for Business]
    O --> B[Authorized Meta Business Portfolio]
    B --> A[Authorized ad account]
    M -->|Permission-scoped operations| API[Meta ads services / Marketing API where applicable]
    API --> A
    A -->|Campaigns, insights, or action outcome| M
    M -->|Structured tool result| C
    C --> R[Report, recommendation, or confirmation request]
Code language: HTML, XML (xml)

The request/response sequence

1. You type: “Show ad spend for my campaigns during October.”
2. ChatGPT determines whether a connected Meta tool can answer that request.
3. The MCP client calls the tool with an account identifier, date range, and relevant parameters.
4. Meta checks your identity, granted permissions, and any account-level rules.
5. The MCP server returns data or an error; ChatGPT formats the findings.
6. For a change request, the server may expose a write tool, but ChatGPT can request confirmation and Meta may block the action. Always validate the final state in Ads Manager.

An OAuth access token is issued after authorization, not a password to paste into chat. MCP calls structured tools rather than scraping the Ads Manager website. The hosted server needs no local installation.

4. Step-by-step: Connect Meta Ads to ChatGPT

The primary instructions below follow OpenAI’s current Plugins experience, rather than the older custom-app menu. The exact browser labels can vary as the products evolve.

Step 1: Prepare Meta Business Portfolio and ad account permissions

Objective: Confirm that your Meta login can access the correct business and advertising account.

1. Open Meta Business Suite and sign in with the Facebook account or Meta Managed Account used for the business.
2. In Business Suite, check Settings to find the intended Business Portfolio and people/asset assignments. You might need an administrator to grant access.
3. Open Meta Ads Manager and select the intended advertising account.
4. Record the ad account ID, account currency, account time zone, and whether the account shows active campaigns. The numeric ID may appear with the act_ prefix in some API contexts.
5. Request the lowest necessary permission: view/reporting for analysis, management rights only when you intend to change campaigns.
6. Where available, open Business Settings > Integrations > Ads MCP server. The portfolio owner can block campaign creation, budget changes, or status edits; these rules are still rolling out. Rules interface.

Inputs: Meta login, intended Business Portfolio, ad account ID, permission level.
Expected result: Ads Manager displays the correct account and campaigns.
If it fails: Ask the portfolio owner to check asset assignments; business access does not necessarily mean ad-account access.

Screenshot placeholder 1: Actual Business Suite portfolio / ad-account permissions screen. Blur account IDs, personal details, and payment methods.

Step 2: Find custom MCP server settings in ChatGPT

Objective: Open the connection creator.

1. Sign into ChatGPT in a desktop browser and select the intended workspace.
2. Open Plugins in the sidebar.
3. Select + (plus) and then Add custom MCP server.
4. If the menu is absent, check the workspace’s permission Create plugins with MCPs, your plan, and your admin/owner role. Ask an admin to create the connection for the workspace if needed.

Expected screen: A form for name, endpoint, authentication, and risk acknowledgment. The new flow does not require Developer Mode. OpenAI guide

Screenshot placeholder 2: Actual ChatGPT Plugins menu with “Add custom MCP server”. Capture in your own workspace; do not use a fabricated screenshot.

Legacy fallback: Older interfaces use Workspace settings > Apps > Create or Settings > Apps > Create, followed by Scan Tools and OAuth.

Step 3: Add Meta’s official MCP endpoint

Objective: Tell ChatGPT which remote server to connect.

1. Paste the exact URL into the endpoint field.
2. Select the authentication option indicated by the server and ChatGPT; the ordinary Meta sign-in path uses OAuth.
3. Read the risk warning carefully and check I understand and want to continue only if your organization approves.
4. Select Create as a plugin.
5. Review the tools and metadata ChatGPT discovers. Do not assume all tools are available simply because the URL was accepted.

Expected result: A connection that exposes discoverable Meta tools.
If it fails: Verify the URL, role, transport, and OAuth setup. Some Meta/OAuth clients have reported registration or issuer errors; a registered OAuth client ID may be required for custom developer-app flows. Browser GET may return HTTP 405 even when MCP requests work.

Screenshot placeholder 3: Actual completed MCP connection form, with authentication details but no credentials exposed.

Step 4: Sign in to Meta using OAuth

Objective: Grant permission without sharing your Facebook password with ChatGPT.

1. When ChatGPT offers to Connect, Authorize, or continue with the sign-in flow, open the Meta authentication page.
2. Confirm the browser is on a legitimate facebook.com / Meta-controlled sign-in or consent page, not an unfamiliar clone.
3. Sign in and complete two-factor authentication if required.
4. Review the information and actions the connection requests. Read-only reporting and campaign management should not be treated as equivalent.
5. Approve only the assets and access you intend to authorize; finish the flow and return to ChatGPT.

Expected result: The connection reports that Meta sign-in is complete and that tools can be discovered or invoked. Never paste a Facebook password, access token, app secret, or backup authentication code into the chat. For custom developer apps, OAuth client IDs, callback registration, token scopes, and App Review are a separate advanced integration path.

Screenshot placeholder 4: Real Meta OAuth consent page; cover personal names and business assets.

Step 5: Verify which advertising accounts are authorized

Objective: Ensure that the connection is scoped to the business you meant to connect.

Meta may offer asset selectors during consent; a dedicated ad-account picker is not guaranteed.

1. In the Meta authorization screens, select only the intended portfolios and assets if those choices are offered.
2. Return to ChatGPT and request the list of available advertising accounts.
3. Compare the returned IDs and names against the account you recorded in Step 1.
4. If an unexpected client or business account appears, stop and reduce the authorization scope before continuing.

Expected result: The correct account is visible. If it fails: Check asset assignments and Business Integrations, then reauthorize as needed.

Step 6: Install and activate the plugin in a conversation

Objective: Make the MCP tools available to the current request.

1. Install the custom plugin created in Step 3, if ChatGPT presents an installation action.
2. Open a new conversation, type @, and select the Meta Ads plugin.
3. Begin with a read-only request; mention the ad account explicitly if you manage multiple accounts.
4. When asking for fresh data in a later message, select or @mention the app again as needed. An app’s selection can apply to the individual message, not necessarily every future message.

Expected result: ChatGPT invokes a real Meta tool rather than guessing performance data. If it fails: Confirm the plugin is installed, accessible in the selected workspace, and enabled for the current request.

Screenshot placeholder 5: Actual ChatGPT composer with the Meta Ads plugin selected.

Step 7: Verify the connection with a low-risk test

Objective: Confirm the integration returns genuine, account-scoped information.

Paste this test:

Using the connected Meta Ads tools, list the advertising accounts I can access.
For each account show name, account ID, currency, and time zone if available.
Do not create, edit, pause, activate, or delete anything.
If the tools cannot access an account, report the error instead of guessing.
Code language: PHP (php)

Next, compare a seven-day spend report with Ads Manager using identical dates, time zone, filters, attribution, and currency.

Expected result: Identifiable accounts and real metrics with a clear reporting period. If the result contains fabricated-looking data or vague general advice: Ask ChatGPT to identify which tool returned each metric, and verify the result manually.

Screenshot placeholder 6: Actual campaign summary in ChatGPT and the matching Ads Manager date range; obscure private campaign names if publishing publicly.

5. Ten-plus ready-to-copy prompts to test the integration

These examples request supported analysis; they do not guarantee that every permission, breakdown, or metric is exposed in your specific connection. Replace names, IDs, and date ranges as needed.

Prompt 1 – List ad accounts

List every Meta advertising account available through this connection.
Show account name, ID, currency, and time zone. Do not make changes.
Code language: PHP (php)

Prompt 2 – Find active campaigns

For ad account [ACCOUNT_ID], list campaigns currently marked active.
Show objective, campaign status, effective delivery status if available,
and budget type. Distinguish configured status from actual delivery.
Code language: JavaScript (javascript)

Prompt 3 – Campaign performance

For [ACCOUNT_ID], summarize campaign performance from 2026-10-01
through 2026-10-30: spend, impressions, reach, clicks, CTR, CPC,
results, and cost per result. Specify any unavailable metrics.
Code language: JavaScript (javascript)

Prompt 4 – Advertising expenditure

Analyze daily Meta ad spend for [ACCOUNT_ID] over the last 30 completed
days. Show the five highest-spend campaigns and identify unusually high
spending days. Explain whether the pattern could reflect budget changes.
Code language: CSS (css)

Prompt 5 – High CPC

Rank campaigns in [ACCOUNT_ID] by link CPC for the last 14 completed
days. Exclude campaigns with fewer than 1,000 impressions from the main
ranking, but list them separately. Include spend and link clicks.
Code language: JavaScript (javascript)

Prompt 6 – Facebook versus Instagram

Compare Facebook and Instagram placements for [ACCOUNT_ID] in the
last 30 completed days. Report spend, impressions, link clicks, link
CTR, cost per result, and attributed conversions where supported.
Flag metrics that cannot be fairly compared across placements.
Code language: CSS (css)

Prompt 7 – Conversion rate

For [ACCOUNT_ID], compare landing-page-view-to-lead conversion rates
by campaign over the last 30 completed days, if both events are
available. State the precise numerator and denominator used.
Do not substitute clicks for landing page views without telling me.
Code language: CSS (css)

Prompt 8 – CPA and ROAS

Calculate cost per acquisition and purchase ROAS for [ACCOUNT_ID]
for the last completed calendar month. Use spend, purchase counts,
and attributed purchase value if available. State the attribution
window, currency, and any missing data. Do not invent revenue.
Code language: JavaScript (javascript)

Prompt 9 – Underperforming ads

Find ads in [ACCOUNT_ID] with meaningful spend but poor results
in the last 14 completed days. Compare each against others sharing
the same objective. Give evidence-based hypotheses, not automatic
recommendations to pause them. Make no changes.
Code language: JavaScript (javascript)

Prompt 10 – Weekly report

Generate a weekly executive report for [ACCOUNT_ID] covering the last
seven completed days versus the preceding seven. Include spend,
results, cost per result, CTR, CPC, top three improvements, top three
risks, and three experiments. Show absolute and percentage changes.
Code language: CSS (css)

6. What can ChatGPT manage? Read versus write

Meta groups tools into reporting, ads, catalogs, signals, help, experiments, and activity logs. The live inventory varies during beta; inspect discovered tools rather than relying on a fixed count. Tool overview

Why “create paused” does not make all editing safe

October 2026 reviews report new entities are generally paused or staged as drafts. Editing an already-running budget, audience, or status is different and may affect live delivery. Verify actual changes in Ads Manager; never assume every edit requires approval. Meta launch | Tool analysis

Example: safely request campaign creation

Design a proposed Meta lead-generation campaign for [ACCOUNT_ID].
Use a daily budget ceiling of [AMOUNT] [CURRENCY] and targeting we
have explicitly approved. First show the complete configuration.
Do not create anything until I approve the plan. If I approve and
the connector supports creation, create it PAUSED. Never activate it.
Return all created object IDs and the final status for review.
Code language: CSS (css)

Before changing live budgets, require approval and verified caps; otherwise edit in Ads Manager.

7. Practical AI-powered marketing workflows

Workflow A: Find wasted budget

Compare campaigns with meaningful spend and high cost per qualified result, grouped by objective. For example, $300/30 leads is $10 CPA, while $450/15 leads is $30 CPA; those numbers alone do not establish lead quality.

Workflow B: Produce a weekly management report

Use Prompt 10 to compare two weeks with absolute and percentage differences; validate unusual results in Ads Manager.

Workflow C: Find promising creative concepts

Compare ads sharing an objective, with enough delivery for a fair comparison. Request CTR, CPC, CPA, and frequency. High CTR does not necessarily mean profitable conversions.

Workflow D: Compare Facebook and Instagram

Compare placement breakdowns using cost per qualified lead, not just impressions. Some conversion totals may not be additive across placements.

Workflow E: Investigate delivery failures

Investigate zero spend by checking status, schedule, review, budget, billing, and tracking. Confirm causes in returned diagnostics or Ads Manager.

Workflow F: Recommend budget changes without executing them

Request proposed budget, percentage change, rationale, confidence, and downside risk. Review business constraints before approving; forecasts are not guarantees.

Workflow G: Understand pixel or dataset quality

Use signal-health tools to identify missing or duplicated events and attribution issues. Confirm findings in Events Manager or a Conversions API audit.

8. Metrics that ChatGPT must calculate correctly

Worked example (illustrative, not retrieved account data): If a campaign spends $200, receives 100 link clicks and 10 attributed leads, link CPC is $2.00 and lead CPA is $20.00. If it records $600 in attributed purchase value, purchase ROAS is 3.0x. Do not calculate ROAS from lead counts alone.

Always align currency, time zone, date range, conversion definitions, filters, and attribution settings; late conversions can change results.

9. Troubleshooting connection and data errors

Diagnose transport, authentication, discovery, Meta permissions, and business errors separately. Do not use repeated write attempts as a connectivity test.

10. Security checklist and change approvals

A connected advertising account gives an assistant access to business-sensitive metrics and potentially real spending controls. Apply defense in depth:

1. Least privilege: Start with read-only permissions. Elevate only the specific users and operations that need changes.
2. Portfolio rules: Where available, restrict ad creation, budget editing, delivery changes, targeting, and creative edits through Business Settings > Integrations > Ads MCP server. Treat absence of this control as a reason for extra caution, not implied approval.
3. Human authorization: Require the assistant to preview intended changes, list affected object IDs and budgets, and wait for explicit approval. Do not assume a natural-language instruction replaces platform approval.
4. Spending guardrails: Use clear per-account budget caps, approved currencies, and maximum change percentages. Verify that any technical rule actually enforces them.
5. Protect OAuth tokens: Never send credentials in chats, tickets, screenshots, prompt templates, or source repositories.
6. Audit and consent: Check activity history after changes, obtain client permission, and avoid blanket access to future business assets.
7. Review untrusted content: Verify AI recommendations and disregard instructions embedded in ad names or retrieved documents (prompt injection).
8. Revoke access: Disconnect the plugin and review/revoke its Meta Business Integration authorization when no longer needed.

OpenAI warns that custom MCP servers can expose sensitive data and write actions. Verify the server and your authorization scope. OpenAI security guidance

11. Limitations and what is not guaranteed

• Beta and account rollout: Not every advertising account sees every Meta tool or rules setting.
• ChatGPT access: Custom MCP creation and write access depend on plan, role, workspace rules, and client surface. Mobile is not the documented custom-connection creation experience.
• Authentication interoperability: Provider OAuth requirements may fail with some MCP clients or custom app configurations. A working URL alone does not prove successful sign-in.
• Not every metric exists: Campaign objectives, tracking quality, privacy thresholds, and attribution determine what can be returned.
• Tool inventories evolve: The number and names of exposed tools may change. Inspect the live discovered list instead of relying on a tutorial’s hardcoded count.
• Rate limits: Meta may throttle usage; no universally verified numeric MCP quota is offered here.
• Read versus write: A server feature may still be blocked by your plan, role, or account rules.
• Approvals: A write tool is not authorization to run spending changes without review.
• Costs: Meta has offered the connector in open beta, but this guide cannot promise permanent free pricing. Advertising spend and any ChatGPT subscription or external agency charges are separate.
• Outside scope: Creative production, CRM validation, and policy disputes may require other systems.

12. Frequently asked questions

1. Can ChatGPT connect directly to Meta Ads?

Yes, with MCP eligibility, Meta authentication, and suitable ad-account rights.

2. What is the official Meta Ads MCP URL?

3. Is Meta Ads MCP free?

Meta launched it in open beta without a stated separate connector fee. Permanent free access is not guaranteed; ChatGPT subscriptions and ad spend are separate.

4. Do I need a Meta Developer account?

Normally no for the standard hosted authorization flow to your own accounts. A developer app may be necessary for advanced custom integration or a commercial app serving other businesses.

5. Do I need a Meta Business Portfolio?

You need access to advertising assets under an appropriate Meta business/account structure; a Business Portfolio is the standard way to manage permissions and policies.

6. Which ChatGPT plan supports full custom MCP access?

Business and Enterprise/Edu have full custom MCP support; Pro is documented for read/fetch. Workspace permissions apply.

7. Can ChatGPT read campaign performance?

Yes if the connection successfully authenticates and the relevant reporting tools, account permissions, and data are available.

8. Can ChatGPT create Facebook or Instagram ads?

Yes, where creation tools and write permissions are available. New entities are generally paused or staged; verify before launch.

9. Can ChatGPT pause a campaign?

Potentially, using a permitted write tool. Do not assume this capability is available on read-only plans or authorizations.

10. Can ChatGPT change advertising budgets?

Potentially, but that is a sensitive write action. Require explicit approval and enforce Meta account-level rules where supported.

11. Is OAuth authentication required?

The normal hosted connection uses Meta sign-in/OAuth. Advanced developer setups can have different supported token configurations; never share credentials in a chat message.

12. Is the integration secure?

It can be used responsibly with trusted endpoints, scoped authorizations, permissions, spend controls, approvals, and audits. Connecting an account with broad write access creates real operational risk.

13. Conclusion

Meta’s official hosted Ads MCP server offers a practical way to ask ChatGPT meaningful questions about authorized Facebook and Instagram advertising accounts. For beginners, the strongest initial use cases are campaign reporting, anomaly investigation, placement comparisons, and executive summaries. Those tasks provide value without immediately risking budget or delivery.

Start with the current Plugins > + > Add custom MCP server route, authenticate through Meta, confirm the exact advertising account, and reconcile the first report with Ads Manager. Only after proving the data is reliable should you consider creating paused campaign drafts or making approved edits. Keep access narrow and verify every live change.

Sources and further reading

Primary documentation and Meta publications

1. OpenAI Help Center: Connecting custom MCP servers in ChatGPT – current setup, plan eligibility, permissions, OAuth considerations, and use.
2. OpenAI Developers: Connect and test your plugin – tool discovery, testing, and troubleshooting.
3. Meta for Developers: Ads MCP server overview – official endpoint and advertised tool families; may require authentication.
4. Meta for Developers: Ads MCP server getting started – Meta connection guidance; may require authentication.
5. Meta Business Help Center: Manage ads from an AI agent – Meta account, authorization, and tools; may require authentication.
6. Meta for Business: Introducing Meta Ads AI Connectors – original product announcement and updates; may require authentication.
7. Meta Newsroom: Meta Ads AI Connectors in open beta – accessible first-party confirmation of open-beta availability.

Supplementary independent reporting used where Meta pages were login-restricted

8. Jon Loomer Digital: Meta Ads AI Connectors Get More Security Controls (August 2026) – hands-on examination of portfolio-level rules.
9. Tofu Ads: What is Meta’s Ads MCP server? (October 26, 2026) – contemporary review of tool categories and sign-in behavior; vendor-authored, not official documentation.

Suggested internal links: Meta Ads Manager beginner guide; Meta Pixel and Conversions API; Facebook Ads not spending; MCP fundamentals; calculating ROAS.

FAQ structured data: Mark up only visible FAQ content; follow current Google eligibility guidance and do not promise rich results.

Rajesh Kumar

I’m Rajesh Kumar, a DevOps, SRE, DevSecOps, Cloud, and Platform Engineering expert passionate about sharing practical knowledge, real-world experiences, and industry best practices. I have worked at Cotocus and regularly write about technology, travel, investing, health, product reviews, and digital marketing through my various platforms.

I publish technical articles at DevOps School, travel stories at Holiday Landmark, stock market insights at Stocks Mantra, health and fitness guidance at My Medic Plus, product reviews at TrueReviewNow, and SEO and digital marketing strategies at Wizbrand.