MCP Ads connects AI agents to Google Ads, Meta Ads, TikTok Ads, Amazon Ads, LinkedIn Ads, Reddit Ads, Microsoft Advertising, Pinterest, Snapchat, X Ads, and more, enabling campaign management, ad performance analysis, targeting, creative management, and reporting through one MCP server.
Encrypted at rest, isolated from the model
Resolved from an AES-256-GCM vault at the moment of the call and attached to the request — the model never sees the secrets.
Try asking
Start here for a campaign launch from images/videos and a brief. Returns supported native build/change tools and platform restrictions in priority order: Meta, TikTok, Google Ads (Demand Gen first), then the other sources. Use the user's brief and supplied media to prepare the complete campaign, groups, creatives and ads with native tool schemas from get_launcher_options/describe_tool. ChatGPT attachments use top-level files and matching asset.fileId; other hosts use accessible trusted URLs or actual bytes. Never invent attachment URLs or transcribe media bytes from visual context. Resolve missing budget, destination, targeting and identities in chat. Never invent provider IDs. Encode explicit user constraints in requirements and identify field origins in provenance. Resolve common carousel settings into each native card, retaining explicit exceptions and order. Exclude DPA/catalog-only ads; collections need explicit image/video plus products. Prepare every requested platform together. Creation keeps delivery paused; activation is a separate explicit option. Pinterest Performance+ rejects PAUSED groups: obtain explicit approval before using performancePlusUnderPausedCampaign:true, which creates an ACTIVE group only under a verified PAUSED campaign, with ads still PAUSED. Never silently relax a strict all-levels-PAUSED requirement. Approval is required after every revision.
Reopen the saved campaign proposal, exact approved revision and per-platform execution results in the MCP App. No advertising calls. Unknown outcomes must be reconciled before any replacement creation.
Stop future work on this exact saved launch. Does not delete or activate native objects. In-flight requests may finish; their receipts remain saved. Releases only unused quota, retaining uncertain outcomes. Requires explicit confirm=true; cannot be undone by a continuation.
List this user's saved proposals and approvals in this workspace. Open a proposal to inspect progress or resume it without duplicate creation.
Read the import saved by this user/workspace’s optional web media picker. Returns ready media IDs and pending/failed files for continuing the same conversation. Does not create ads or import another copy.
Open an in-conversation Google Drive gallery. Start here to add new Drive images/videos: supply a connectionId and a new requestId, offer the Google selection button, then call again with the SAME IDs after the user returns. Multiple files can be selected together. Only explicitly picked files are authorized; a folder link is not a grant. Selected files return preselected, without copying a prompt. No import or ad creation occurs until requested.
Browse a page of existing Meta, TikTok, Google Ads or Snapchat image/video media in the selected account with visual previews. This opens a selection picker: wait for the user to choose; listing results is not a selection. If the user already selected exact media in the conversation, reuse those references rather than opening a new library picker. Reuse the returned native object in a proposal asset with field nativeId; selected native files count toward the lot limit and consume no incoming volume. No R2 copy or ad creation. Use cursor for the same account and kind. Native IDs cannot move between accounts or platforms; request an accessible original for another destination.
Disconnect this workspace Drive/Dropbox source and remove stored credentials. Owner/admin only; explicit confirm=true required. Existing cloud originals and native ads are untouched. Temporary copies keep their normal expiry.
Cancel unfinished transfers in this user/workspace import and release remaining reservations. Requires explicit confirm=true. Completed files retain their expiry and existing advertising objects are untouched.
View one provider-generated thumbnail for a selected Drive/Dropbox file version. Call before making visual judgments. Does not import an original or inspect every video frame. If unavailable, report it rather than infer the visual from its name.
Read explicitly saved naming, UTM and stable targeting conventions for a named client in this workspace. These settings never supply budgets/dates or authorize ads. Mark inherited values as client_settings in the launch provenance.
List connected Google Drive and Dropbox sources in this workspace, plus connection links. For adding Drive files in the conversation, call open_launcher_drive_picker with a connectionId and new requestId. Does not import files or grant access. Cloud sources are shared by this workspace.
Read this user/workspace’s saved Drive or Dropbox selection. Import the exact returned source objects with import_launcher_media; reuse importRequestId on retries. No ads are authorized.
Read the exact file versions explicitly selected in Google Picker by this user/workspace. Use the returned source objects with import_launcher_media and the returned importRequestId on retries. Selection grants no ad creation permission.
Read one page of a user-specified Dropbox folder, or already authorized Drive files within a folder. Drive requires Google Picker selection first: a folder link never grants access to children. For new Drive files prefer open_launcher_drive_picker with this connectionId and a new requestId; refresh the same request after selection to keep the gallery in the conversation. Use the returned cursor for the same folder; open subfolders only when requested. Returns names, types, sizes and immutable source versions. Metadata is not visual inspection. No original files are copied.
Read saved import progress and ready media IDs. Reference ready IDs as asset.mediaId in a campaign proposal. Failed files and expiry remain visible; never silently omit them.
Read Launcher limits and current reserved/used volume and ad allowance. Sources and workspaces in the same subscribing organization share these quotas. Return limits, renewal date and billing link; never charge or upgrade automatically.
Complete current capability inventory for any or all sources, including tools omitted from tools/list. MUST call before answering 'what can you do?' or saying a feature is unavailable. Includes configuration fields and access restrictions. Covers TikTok, Meta, Google Ads, Amazon Ads, Pinterest, Snapchat, LinkedIn, Reddit, Microsoft Advertising, GA4 and Search Console; do not infer one source's capabilities from another.
Start here for any Google Sheets export or automatic sync request, including hourly refresh. Discover accounts, the full source reporting catalog, Google connections and plan limits. Optional source narrows the listing; include_field_details=true returns native field mappings, formats and compatibility constraints. Meta includes purchases, purchase_value, cpa and roas; purchase_event selects omni or website attribution. Never claim Sheets only supports delivery metrics. Immediately call prepare_sheet_export directly next, using the user's known choices; its compact MCP App card collects destination/refresh settings and confirms activation. Resolve missing source, accounts, metrics, dimensions and filters in the conversation, asking only for genuinely missing or ambiguous data choices. Do not substitute the host's generic questionnaire or ask again for specified metrics/frequency. Sheets exports execute in the Data backend without an assistant staying open. Only owners/admins can manage exports.
Create the interactive Google Sheets export card (MCP App) from the user request. Call this tool directly so the host renders the card; do not replace it with generic questions or a Cowork/Apps Script scheduling choice. Supply a UUID request_id and the report inferred from the request. Keep report scope and filters in the conversation; the card shows them read-only and cannot choose another source/account or build filters. Resolve missing or ambiguous data choices conversationally. Period, destination or Google connection may be omitted and completed in the compact card. Prefill every known choice; do not ask the user to enter it twice. Metrics and dimensions come from get_sheet_export_options. filters are AND; filter_groups are OR groups of AND conditions, combined with filters. With date, filters apply to daily rows; omit date for native period totals per entity, then filter those totals. Ratios and reach are never summed across days. Do not silently substitute unsupported metrics, omit unsupported filters, or interpret ambiguous spend thresholds. Validate against the supported catalog; explain an unsupported request and agree on a supported alternative before preparing the export. No Google file is created until start_sheet_export. Use schedule.type=hourly when hourly refresh is requested (Team/Enterprise); period.type=today follows the selected timezone on every run. dimensions=[date,hour] requests hourly Meta/TikTok rows; it is independent of refresh frequency. For multiple sources use separate exports/tabs in the same spreadsheet: start the first reviewed export then reuse its spreadsheet ID and Google connection for the next source. Defaults to one-off only when no recurring schedule was requested.
Show the interactive Google Sheets export card directly in the conversation. Call this tool directly, not through run_tool, when available. Open the saved Sheets export, missing choices, current revision and execution history. Renders the compact export MCP App where supported; returns the same structured configuration for text clients. No extraction is triggered. After external Google connection/file selection, reopen this export.
Save the full revised Sheets export configuration using its current revision. Existing exports are paused after edits and must be started again explicitly. Running jobs prevent edits. Never broaden accounts or replace an existing tab unless requested. Set target.replace_existing_tab only after the user has chosen to replace that exact tab.
Start the exact reviewed Sheets export revision with confirm=true after the user requests creation/activation. Creates or reuses its destination, verifies Google edit access, and launches the first backend job. Repeated calls for the same active revision do not create another file/job. Requires report, period and Google connection; use connect_url from open_sheet_export when needed. After success, immediately reply with the clickable spreadsheet_url and actual job status; queued does not mean data has finished loading. Continue preparing any other sources explicitly requested by the user in distinct tabs of the same spreadsheet, checking existing exports to avoid duplicates; do not activate other exports without confirmation. All values in the selected dedicated tab will be refreshed; other tabs are preserved.
List saved Sheets exports and their latest execution state, next run, file links and actionable errors. No source extraction. Use to identify an export before modifying, pausing or running it.
Pause, resume, run once or archive the selected Sheets export when requested. Requires current revision and a UUID request_id for retries. Pause/archive cancel pending executions; archive does not delete the spreadsheet. Run once does not activate a paused schedule. Native hourly/daily/weekly backend jobs continue after the conversation closes; no Cowork task or Apps Script is required.
List the workspaces explicitly authorized for this connection and their selected accounts, with live permissions. Call list_accounts to resolve the user's account or client/workspace by name. It returns only authorized workspaces and selected accounts, grouped by workspace with IDs and permissions. If the named account is unique, call it directly with its workspace_id; if a workspace is named, use only its selected accounts for the requested source. If only one account matches the source, proceed without asking. Otherwise ask ONE clarification with accounts grouped by workspace, offering all accounts in a workspace as an option. Reuse an explicitly chosen workspace in this conversation and always label results with the workspace and account. Never silently combine clients on a vague request. Cross-workspace comparisons require an explicit request: call each workspace separately and keep currencies/attribution separate. Pass workspace_id on every follow-up, including saved reviews and exports. Workspace selection is per call, never shared mutable session state. Adding workspaces requires new OAuth consent; changing the dashboard workspace does not change an existing grant.
Accounts this organisation has explicitly allowed, per platform. This is not everything the connections can reach: only accounts ticked in the dashboard are accessible, and any other identifier is refused. Each workspace carries its shared client memory (clientMemory): apply it. Call list_accounts to resolve the user's account or client/workspace by name. It returns only authorized workspaces and selected accounts, grouped by workspace with IDs and permissions. If the named account is unique, call it directly with its workspace_id; if a workspace is named, use only its selected accounts for the requested source. If only one account matches the source, proceed without asking. Otherwise ask ONE clarification with accounts grouped by workspace, offering all accounts in a workspace as an option. Reuse an explicitly chosen workspace in this conversation and always label results with the workspace and account. Never silently combine clients on a vague request. Cross-workspace comparisons require an explicit request: call each workspace separately and keep currencies/attribution separate. Pass workspace_id on every follow-up, including saved reviews and exports. Workspace selection is per call, never shared mutable session state. Adding workspaces requires new OAuth consent; changing the dashboard workspace does not change an existing grant. Preserve the user-selected media and account throughout campaign preparation. If a media picker is opened, wait for its selection handoff; do not independently select other assets from a second gallery. Do not infer permission to reuse a brand identity from its presence on an image; if the creative/identity and destination refer to different businesses, ask for clarification. For campaigns from a brief and media, use get_launcher_options then prepare_campaign_launch directly to show the complete campaign proposal/MCP App. New entities are paused by default. Ask for approval of the displayed revision before publish_campaign_launch; enable activation only when explicitly requested. Multi-platform requests belong in one proposal. Continue approved launches using saved checkpoints; never recreate uncertain writes. Before answering what you can do, or claiming a capability is absent, call get_capabilities for the platform. tools/list is only a starter selection, never the complete capability inventory. Advanced creation, targeting/configuration edits, uploads and renaming may be discoverable only. Use the returned real tool names and schemas; distinguish implemented capabilities from plan, connection, endpoint and provider restrictions. For Google Sheets exports and hourly sync, use get_sheet_export_options then prepare_sheet_export directly to render the MCP App; Get MCP Ads runs refreshes natively after the conversation closes. No Cowork task, Apps Script or user API key is needed. A catalog entry is not proof of a successful provider call. Never claim every format or objective is supported.
Return the full input schema and platform of any tool on the gateway, including tools not listed here. Call this before run_tool so the arguments are right.
Execute any tool on the gateway by name, including those not listed here. Use search_tools to find one and describe_tool to learn its arguments. Account and plan checks apply exactly as they do to the listed tools.
Search the current server catalog, including newly deployed tools, creative performance galleries and platforms not yet connected. Call whenever you need a capability or a cached tool is missing. Space-separated keywords are matched independently. Preserve the user-selected media and account throughout campaign preparation. If a media picker is opened, wait for its selection handoff; do not independently select other assets from a second gallery. Do not infer permission to reuse a brand identity from its presence on an image; if the creative/identity and destination refer to different businesses, ask for clarification. For campaigns from a brief and media, use get_launcher_options then prepare_campaign_launch directly to show the complete campaign proposal/MCP App. New entities are paused by default. Ask for approval of the displayed revision before publish_campaign_launch; enable activation only when explicitly requested. Multi-platform requests belong in one proposal. Continue approved launches using saved checkpoints; never recreate uncertain writes. Before answering what you can do, or claiming a capability is absent, call get_capabilities for the platform. tools/list is only a starter selection, never the complete capability inventory. Advanced creation, targeting/configuration edits, uploads and renaming may be discoverable only. Use the returned real tool names and schemas; distinguish implemented capabilities from plan, connection, endpoint and provider restrictions. For Google Sheets exports and hourly sync, use get_sheet_export_options then prepare_sheet_export directly to render the MCP App; Get MCP Ads runs refreshes natively after the conversation closes. No Cowork task, Apps Script or user API key is needed. A catalog entry is not proof of a successful provider call. Never claim every format or objective is supported.
Search the operating skills: written procedures for media-buyer jobs (audits, pacing, fatigue, incidents), each grounded in real tools with decision rules and failure modes. Search by need in plain words, e.g. 'creative fatigue' or 'why did spend drop'. Then call get_skill for the full contract before executing.
Return one skill's full contract: when to use it and when not to, the tool sequence, the decision rules, the evidence requirements, the human approval boundary and the known failure modes. Read it before executing the skill.
Open the Get MCP Ads cockpit: every client workspace and selected ad account, triggered alerts, ChatGPT automations, campaign launches, Sheets exports and saved reviews in one place, with per-account performance loaded on demand. Use when the person asks to see their accounts, clients or overall ads status, to compare their accounts' spend, results and cost per result, or to open the cockpit. Reads Get MCP Ads records; the app loads each account's figures from its platform, next to the previous period.
Open the Ads panel beside the conversation: one ad account's KPIs, campaigns and creatives that stay visible while chatting. Selecting rows or creatives shares them with ChatGPT as context. Opened by the person from the conversation panel; pass platform and account_id to focus an account.
Account totals and campaign rows for one selected ad account over a period, with the equal-length previous period. Used by the Ads cockpit and panel. Meta, Google Ads, TikTok, Pinterest, Google Analytics and Search Console.
Create, rename or archive a client workspace, move ad accounts between clients, or set a Member's access to a client, with the dashboard's rules (Owners and Admins; creating clients and moving accounts out of the default workspace depend on the plan). Used by the Ads cockpit.
What was changed in the organisation's ad accounts and Get MCP Ads, by whom and when, newest first: the web app's audit log. Owners and Admins only; kept 7, 30 or 365 days by plan. Used by the Ads cockpit's Activity.
The breakdowns the Ads cockpit can show for one selected account: for Google Analytics, the default channel group, the property's custom channel groups, source / medium, campaign, landing page, device and country.
One level of a selected ad account (campaigns, ad sets, ad groups or Performance Max asset groups, ads, or Google Shopping products, brands and product types) over a period, with the equal-length previous period and each row's parents. Used by the Ads cockpit's drill-down. Meta, Google Ads and TikTok.
Read the person's Get MCP Ads defaults: default client workspace, default reporting period, KPI focus and previous-period comparison. Apply them only when a request names no client or dates.
Change the person's Get MCP Ads defaults when they ask to: default client, default reporting period, KPI focus, previous-period comparison. Only the given settings change. Never changes advertising accounts.
List the organisation's advertising and analytics sources: which are connected, which need reconnecting, each login of a platform (connections, for those who manage sources), every ad account each connection can reach, which accounts Get MCP Ads may use (ticked), and the plan's account count. Use before connect_source or select_ad_accounts, or when the person asks what is connected. No provider call.
Start connecting an advertising or analytics platform when the person asks: a new platform, another login to a platform already connected (another Business Manager or Google account), or reconnecting one connection that expired. Returns a one-time link to the platform's own consent screen, valid 10 minutes, for this person only. In the Ads cockpit the app opens it; in conversation, give the link as a button or link. After the person consents, call get_sources to see the accounts found, then select_ad_accounts. Owners and admins only.
Set which ad accounts of one connected platform Get MCP Ads may reach, when the person chooses them. Replaces that platform's whole selection: pass every account to keep. Same rules as the dashboard (role, plan account count). It never buys anything; past the plan's count it explains what to untick. Owners and admins only. Nothing changes in the ad accounts themselves.
Pause, reactivate or set the daily budget of up to 20 campaigns or ad sets, or pause and reactivate ads, of one selected account (Meta, Google Ads status only, TikTok, Pinterest). Always two calls: without confirm_token it only returns a preview and a confirm token; show it, and call again with the same arguments and confirm_token only after the person confirms. The token expires and is bound to these exact arguments.
List the organisation's members with their roles, and pending invitations. Use when the person asks who has access or before inviting someone. No provider call.
Invite one person by email to this Get MCP Ads organisation when the person asks: admin, or member with read or write access to chosen client workspaces. Sends the invitation email, which cannot be unsent: confirm the email, role and workspaces with the person before calling. Same rules as the dashboard (role, seats, hourly limit). Owners and admins only; never buys a seat.
Summarize this Get MCP Ads connection: plan, client workspaces, connected advertising sources and the number of selected accounts, with the link to manage them. No provider call.
Return the Get MCP Ads profile represented by this connection's credentials. The opaque id is unique within this app and remains unchanged across token refresh, reconnection, and display-metadata changes.
Typeahead search for composer @mentions: client workspaces, ad accounts, campaigns already seen through Get MCP Ads, alerts and campaign launches.
Ask the person to pick one of their selected ad accounts in a native form with the platform logos, when a request is ambiguous between several accounts or clients. Returns the chosen workspace_id, platform and account_id. Hosts without forms receive the list to ask in chat.
Show the images or videos of an ad account's media library as a visual picker and return the ones the person selects, with their native IDs for a campaign launch (asset field nativeId). Meta, TikTok, Google Ads and Snapchat. Nothing is created or copied. Hosts without native forms receive the library picker App instead.
One endpoint, the same key, whichever client you use.
~/Library/Application Support/Claude/claude_desktop_config.json (Mac) · %APPDATA%\Claude\claude_desktop_config.json (Windows)
Replace API_KEY with your own key.
Already have an "mcpServers" section in your config? Just add the server entry inside it.
Discovery, routing, credentials, tool scoping and execution logs all happen at the gateway→connections stay ACTIVE with no work from you
Get MCP Ads runs through a gateway that holds the credentials, scopes the access and records every call.
Managed auth, hosted MCP servers, and every Gmail tool your agent needs.
Free to start.