Meta Ads MCP Server
Social
Connect any AI assistant to Meta Ads through one secure MCP connection to analyse, create and optimise campaigns — with every spend held for human approval.
What is the Meta Ads MCP?
The Meta Ads MCP Server is an integration that connects AI assistants — ChatGPT, Claude, Claude Code, Cursor and Codex — to your Meta Ads account through a single secure MCP (Model Context Protocol) connection. Instead of clicking through Ads Manager, you describe what you want in plain language and the AI researches audiences, drafts campaigns, interrogates performance across Facebook and Instagram, and prepares optimisations. OpenAds sits in the middle as a control layer, translating natural-language intent into precise Meta Marketing API calls.
It matters because Meta’s placements, audience signals and delivery algorithms move faster than most teams can manually monitor. The AI closes that gap while OpenAds keeps you firmly in charge: analysis, proposal, approval and execution are strictly separated. The assistant can read anything and propose anything, but every action that spends money or changes your account — launching a campaign, shifting budget, editing an audience — is held for explicit human approval before it executes. You get the speed of automation without surrendering control.
Key capabilities
Pull campaign, ad set and ad performance across Facebook, Instagram, Messenger and the Audience Network
Research and size Custom, Lookalike and interest-based audiences
Identify wasted spend by placement, audience overlap and creative fatigue
Draft new campaigns with objectives, budgets, targeting and creative structure
Propose budget reallocation and bid-strategy changes to hit your CPA or ROAS target
Generate stakeholder-ready reports with clear commentary
Compare Advantage+ automated delivery against manually structured campaigns
Hold every write action for human approval before it touches your account
What you can do
Analyse performance
Ask for performance across any window and OpenAds breaks it down by campaign, ad set, placement, device and creative. It surfaces where CPA is drifting, which Instagram versus Facebook placements are carrying delivery, and how frequency is affecting results — with plain-language commentary rather than raw exports.
Research keywords & audiences
Meta is audience-led, so the assistant helps you size and compare Custom Audiences, Lookalikes and detailed-interest segments. It flags audience overlap that inflates costs, suggests exclusions to stop retargeting converters, and recommends seed sources for Lookalikes based on your highest-value events.
Find wasted spend
OpenAds scans for creative fatigue signalled by rising frequency and falling CTR, ad sets stuck in the learning phase, redundant audiences competing in the same auction, and placements delivering impressions but no conversions. Each finding comes with a proposed fix and the estimated spend at stake.
Create campaigns
Describe your objective, budget and audience and the assistant drafts a complete structure — objective selection, Advantage+ or manual ad sets, placements, schedule and creative slots. Nothing goes live until you review the full plan and approve it.
Optimise & pace budget
The assistant tracks pacing against your monthly target and proposes reallocations toward ad sets beating your ROAS goal, away from fatigued creative, and between Advantage+ and manual campaigns. It can recommend bid-strategy shifts, always as a proposal for your sign-off.
Generate reports
Request a weekly, monthly or campaign-level report and OpenAds assembles the numbers with narrative context — what changed, why, and what to do next — ready to paste into a client update or brief a stakeholder without hours of manual formatting.
Supported campaign types
Campaign typePrimary useTypical workflow Sales (conversions)Drive purchases and high-value eventsAnalyse ROAS, propose budget shift, approve, execute LeadsCapture form fills and enquiriesCompare cost per lead by audience, refine targeting Advantage+ ShoppingAutomated e-commerce prospecting and retargetingBenchmark against manual, reallocate budget on approval AwarenessMaximise reach and brand recallTrack frequency and CPM, refresh creative TrafficSend qualified visitors to site or appOptimise placements, prune low-CTR ad sets EngagementGrow interactions, messages and video viewsIdentify top creative, scale winners on approval
Getting started
Sign in to OpenAds and open the integrations directory.
Select Meta Ads and start the secure connection flow.
Authorise access via Meta's login and grant the ad accounts you want to manage.
Choose the specific ad account, or accounts, OpenAds should connect to.
Confirm your approval preferences so write actions are held for sign-off.
Once connected, link your preferred AI client — ChatGPT, Claude, Claude Code, Cursor or Codex — to the OpenAds MCP endpoint. The same secure connection works across every client, so you can research in one and approve in another. Your credentials stay with OpenAds; the assistant only ever sees the scoped, permissioned connection.
First prompts to try
"Show me last 30 days on Meta by campaign, sorted by ROAS."
"Which ad sets have rising frequency and falling CTR? Flag creative fatigue."
"Find audience overlap that's costing me money and suggest exclusions."
"Draft a Sales campaign for our new collection with a ₹2,00,000 monthly budget."
"Compare our Advantage+ Shopping campaign against the manual prospecting set."
"Propose a budget reallocation to hit a ₹450 target CPA."
"Build me a weekly performance report with commentary for the client."
"Which Instagram placements are converting best right now?"
A complete workflow example
Objective: Reduce cost per purchase without cutting total conversions.
Prompt: "Our Meta CPA has crept up this month. Find the causes and propose fixes to bring it back under ₹500."
What OpenAds analysed: the assistant pulled 30 days of performance across every active campaign, broke results down by ad set, placement and creative, and checked frequency, learning-phase status and audience overlap.
Findings: two ad sets showed frequency above 4.0 with CTR down by a third — clear creative fatigue. A Lookalike audience overlapped heavily with an interest set, and the Audience Network placement was spending with almost no conversions.
Proposed actions:
Pause the two fatigued ad sets and shift their ₹80,000 budget to the top-performing Advantage+ set
Add an exclusion to remove the overlapping audience
Turn off the Audience Network placement and redistribute spend
Approval step: OpenAds presented all three changes as a single approval request with the projected CPA impact. Nothing was paused or moved yet — the plan simply waited for a decision.
Result: after review, the changes were approved and executed exactly as proposed. Blended CPA settled back under ₹500 within the week while total purchases held. Every step was visible, reversible and authorised — nothing executed without approval.
Best practices
State your true north metric — target CPA or ROAS — so proposals are ranked against it.
Let ad sets exit the learning phase before judging or restructuring them.
Review audience overlap regularly; it is the most common hidden cost on Meta.
Refresh creative before frequency climbs rather than after performance drops.
Approve changes in batches so you can see the combined projected impact.
Safety and control
OpenAds draws a hard line between read and write actions. Read actions — pulling metrics, sizing audiences, generating reports — run freely so the assistant can research without friction. Write actions — launching campaigns, changing budgets, pausing ad sets, editing audiences — are always held at an approval checkpoint. You see exactly what will change and its projected effect before anything happens, and nothing executes until you approve it. Analysis, proposal, approval and execution stay separate by design.
OpenAds vs manual Meta Ads management
AspectOpenAdsManual management Analysis speedCross-account insight in seconds via natural languageManual filtering and exports in Ads Manager Wasted-spend detectionProactive flags on overlap, fatigue and dead placementsRelies on analyst catching it manually Campaign setupFull structure drafted from a promptBuilt field by field by hand ControlEvery change held for explicit approvalDirect edits, easy to fat-finger ReportingNarrative reports generated on requestHours of manual assembly
Limitations
The assistant proposes changes; it cannot bypass the approval checkpoint or act autonomously on spend.
Creative production is out of scope — OpenAds structures and optimises campaigns but does not generate images or video.
Results depend on clean conversion tracking; weak or missing Pixel and Conversions API signals limit accuracy.
Meta's automated delivery systems still control final placement and auction outcomes within your approved settings.
Frequently asked questions
Can the AI spend my budget without asking?
No. Every action that spends money or changes your account is held at an approval checkpoint. The assistant can analyse and propose, but you approve before anything executes.
Which AI assistants work with the Meta Ads MCP?
ChatGPT, Claude, Claude Code, Cursor and Codex all connect through the same secure OpenAds MCP endpoint. You can research in one client and approve in another.
Does it work across Facebook and Instagram?
Yes. OpenAds covers Facebook, Instagram, Messenger and the Audience Network, and can break performance down by placement so you see where results come from.
Is my Meta login shared with the AI?
No. Authorisation stays with OpenAds through Meta's secure login. The assistant only sees a scoped, permissioned connection — never your raw credentials.
Can I limit which ad accounts it can touch?
Yes. You choose exactly which ad accounts to connect during setup, and you can adjust that scope at any time.
Connect it through one secure MCP connection.