← Blogs / AI & MarTech Strategy

Ad Platforms Now Speak MCP: How to Connect an Agent to Meta, Google, TikTok, Snap and LinkedIn

Santosh Pradhan·October 10, 2026

I'm Santosh Pradhan, a MarTech Solutions Architect in Munich. Over the past months the ad platforms have started publishing their front doors for software agents, and the question I get most is practical: how do we actually connect one?

Short answer: Meta, Google Ads, TikTok and Snap each document an MCP server you can connect an agent to. LinkedIn, as far as I can find, does not; it offers Marketing APIs behind a partner approval. Start read-only, use the platform's own endpoint, and keep your guardrails outside the agent.

What changed in 2026

TikTok introduced its Agentic Hub in July, a marketplace where advertisers run native and third-party AI tools on top of TikTok's MCP server. Partners publishing agents include HubSpot, Wix, Constant Contact, Innovid and Kochava. Pinterest and Snap followed with MCP efforts of their own. Meta said on its January earnings call that agentic shopping tools were coming, and guided 2026 capital spending to $115B to $135B. OpenAI's early ad program was reported at roughly $60 per thousand views against about $20 for Meta, with only basic impression and click reporting.

That last point matters for architecture. An agent can only optimize what it can measure, so channels that don't return attribution data keep the loop open.

Two protocols, two jobs

MCP connects an agent to tools and data. A2A, proposed by Google in April 2025 and donated to the Linux Foundation that June, connects agents to each other as peers. At its one-year mark A2A reported more than 150 supporting organizations and SDKs in Python, JavaScript, Java, Go and .NET. Version 1.0 added JSON-RPC, REST and gRPC bindings, multi-tenancy and signed Agent Cards for identity. They are complements, not competitors.

How to connect an agent, platform by platform

Each platform follows the same shape: a remote endpoint, a sign-in, and permissions that inherit from your account role. Check the linked documentation first, because these details change quickly.

Meta (Facebook and Instagram)

Endpoint: https://mcp.facebook.com/ads. Documented tools cover reporting and insights, campaign, ad set and ad creation and editing, catalog and feed troubleshooting, signal health, Business Help Center search, A/B tests and lift studies, and the account activity log. See Meta's Ads MCP Server documentation.

  1. Confirm you have an advertiser or admin role on the Business Manager that owns the ad account.
  2. In your AI tool, add a custom connector or remote MCP server and paste the endpoint above. In Claude this is under Settings, then Connectors.
  3. Complete the Meta OAuth sign-in. A third-party write-up reports that no developer app or API key is needed.
  4. Select the Business Manager and ad account, and grant the scopes requested. Read access works with lighter scopes.
  5. Test with a read request, such as summarizing active campaigns.
  6. Grant ads_management only when you want the agent to change campaigns. Write actions are live, with no draft mode or one-click undo.
  7. If the tool reports is_ads_mcp_enabled: false, the phased rollout has not reached your account yet. Check again in a few days.

The launch date and sign-in details come from a vendor write-up, so confirm them against Meta's guide.

LinkedIn

No official MCP server was reported as of mid-2026. The route is the Marketing API under api.linkedin.com/rest/, which is partner-gated.

  1. Prepare a registered legal entity, a verified business email and a verified Page super admin.
  2. Create an app in the LinkedIn Developer Portal.
  3. Apply for the Advertising API product, describing a supported commercial use case.
  4. Start on the Development tier, then request Standard once approved.
  5. Implement three-legged OAuth 2.0. Members must consent, and client-credentials tokens are not accepted for Marketing APIs.
  6. Send a LinkedIn-Version header in YYYYMM format on every request.
  7. Plan for re-authorization. Access tokens last about 60 days, and refresh tokens are limited to approved partners.
  8. If you want MCP, wrap this API yourself or use a vendor server built on it. Avoid any server that drives a logged-in browser session, which is reported to violate LinkedIn's User Agreement.

Google Ads

Google's open-source server is read-only for now. Tools: list_accessible_customers, search (GAQL queries) and get_resource_metadata. See Google's MCP server guide.

  1. Create a Google Cloud project and confirm it has Explorer, Basic or Standard Google Ads API access.
  2. Prepare credentials: an OAuth client ID and secret, application default credentials, or a service account.
  3. Add an mcpServers entry to your host's config that runs the server through pipx.
  4. Set GOOGLE_PROJECT_ID and GOOGLE_APPLICATION_CREDENTIALS. For manager accounts, also set GOOGLE_ADS_LOGIN_CUSTOMER_ID.
  5. Restart the host and run list_accessible_customers to confirm access.
  6. For a shared deployment, build the Docker image, deploy it to Cloud Run and point your client at the service's /mcp URL.

Google's page says developer tokens were sunset on 9 September 2026, with access levels now tied to the Cloud project.

TikTok

See TikTok for Business: MCP server.

  1. Choose an endpoint. Full Disclosure, https://business-api.tiktok.com/open_mcp/tt-ads-mcp-flat, loads about 400 tools and suits large context windows.
  2. Or choose Progressive Disclosure, https://business-api.tiktok.com/open_mcp/tt-ads-mcp-layer, which loads about 40 core tools and finds the rest on demand.
  3. Add the endpoint to your AI tool as a remote MCP server. TikTok describes setup as zero-code with no developer credentials.
  4. Follow TikTok's connection guide for the sign-in step, since the overview page does not detail it.
  5. Test with a reporting request before enabling campaign changes.
  6. Optionally add Skills from the Agentic Hub marketplace.

Snap

See MediaPost's report on Snap's Ads MCP.

  1. Sign in as an organization admin in Snap Ads Manager.
  2. Connect your AI tool to the Ads MCP server. Claude, ChatGPT and Gemini are named as examples.
  3. Approve that agent and choose which data it can reach. Each agent is approved separately.
  4. Ask a reporting question, such as a seven-day summary.
  5. Expect read-only access. Snap plans write access later, with admins choosing which agents get it.

Pinterest

Pinterest announced an MCP effort, but I found no primary documentation. Check Pinterest's developer site directly before following any third-party guide.

Principles for enterprise and solution architects

  1. Treat each ad platform as an API product. Build one governed gateway instead of a custom connector per channel.
  2. Keep tool access and agent-to-agent delegation separate. They have different trust models, so give each its own identity, audit trail and rate limits.
  3. Verify who you are talking to. Signed Agent Cards give you something to check before a third-party agent touches budget.
  4. Own your attribution. If a channel returns only impressions and clicks, measurement has to live in your data layer.
  5. Put guardrails outside the agent. Budget caps, approval thresholds and kill switches belong in systems you control.

Checklist that applies everywhere

  1. Start with a read-only connection and a test or low-spend account.
  2. Use the platform's own endpoint. If a tool asks for an API key where the vendor says OAuth only, treat it as unofficial.
  3. Give the agent the narrowest role and scopes that work, and allow write actions only with a human approval step.
  4. Log every agent action in a system you control, and expect reported spend to lag and change for several days.

FAQ

Which ad platforms offer an official MCP server?

Meta, Google Ads, TikTok and Snap each document one. LinkedIn was reported to have none as of mid-2026.

Can an agent change my campaigns?

On Google's server and at Snap's launch, no, both are read-only. Meta and TikTok document campaign changes, with Meta requiring the ads_management scope.

What is the difference between MCP and A2A?

MCP connects an agent to tools and data. A2A connects agents to each other as peers.

Sources

Santosh Pradhan

Santosh Pradhan

MarTech Solutions Architect · Munich