Ads MCP Server
One MCP connection for every ad platform you have connected to Pipeboard — Meta, Google Ads, TikTok, LinkedIn, Pinterest, Reddit, Snapchat and Google Analytics 4. Ask across platforms in plain language, and act where the work is.
https://mcp.pipeboard.co/ads-mcpWhat is the Ads MCP?
The Pipeboard Ads MCP is a Model Context Protocol server that connects AI assistants — Claude, ChatGPT, Cursor, or any MCP client — to the ad platforms you have connected to Pipeboard. It is an aggregator: one URL that re-lists every platform’s tools under a platform-prefixed name, then routes each call to the platform that owns the tool.
That matters when the work spans platforms. “Compare last week’s spend and ROAS across Meta, Google and TikTok” is one question to one connection, rather than three assistants that each see a slice.
The per-platform URLs still work. Each platform also has its own server (meta-ads.mcp.pipeboard.co, and so on), which is the better choice when you deliberately want an assistant scoped to one platform. The unified URL is additive, not a replacement.
Getting Started
Under 2 minutes
1. Sign up for a Pipeboard account:
pipeboard.co/auth/signup →2. Connect at least one ad platform, and choose which ad accounts Pipeboard may reach:
pipeboard.co/connections →Connect as many as you like — every connected platform shows up on the same MCP connection, with no client reconfiguration.
3. Add the MCP server to Claude (or another MCP client):
https://mcp.pipeboard.co/ads-mcpUse this URL as a Streamable HTTP MCP server. Authentication runs over OAuth 2.0 with Dynamic Client Registration, so a DCR-capable client (Claude included) registers itself and takes you through the consent screen — there is nothing to paste. For clients without OAuth, generate an API token in settings and send it as a bearer token.
What one connection covers
Tools appear for the platforms you have connected — nothing else
Facebook & Instagram campaigns, creatives, audiences, catalogs, lead-gen forms
Search, Shopping and Performance Max campaigns, keywords, GAQL reporting
Campaigns, ad groups, creatives, identities, reporting
Campaign groups, campaigns, creatives, audience and demographic reporting
Campaigns, ad groups, pins, audiences, reporting
Campaigns, ad groups, ads, audiences, reporting
Campaigns, ad squads, creatives, audiences, reporting
Reports, funnels and conversion data to check what the ads actually drove
Two more platforms are wired in but listed only for accounts that have been enabled for them:
Tool names, and finding the one you need
Because one connection serves several platforms, every tool is namespaced by its platform. The same reporting call reads meta_ads_get_insights on Meta and google_ads_get_campaign_metrics on Google Ads, so an assistant can never act on the wrong platform by accident.
The full surface is several hundred tools. Rather than reasoning over all of them, an assistant can call tool_search with a plain description of the task and get back the best-matching tool definitions, ranked:
- Every tool carries a human-readable title, so your client shows "Get Campaign Metrics", not get_campaign_metrics.
- Every tool carries a read-only or write annotation, so reads and writes are distinguishable before you approve a call.
- manage_account_slots is listed once, un-prefixed — account slots are platform-agnostic.
Access, permissions and safety
Access is explicit and account-scoped. Three independent limits apply to every call, and all three fail closed — a call that is not clearly permitted is refused.
Which platforms and accounts
Only the platforms you connected, and within those only the ad accounts you selected under Configure Access. A platform you have not connected returns connect guidance instead of data; an account outside the selection is refused.
Read-only or read-write
Each API token carries its own scope, enforced server-side on every call. A read-only token cannot call a write tool whatever the prompt says, and read-only team members are limited the same way — write tools are not even listed for them.
Nothing spends unless you ask for it
New campaigns are created paused by default on Meta, Google Ads, Reddit and Snapchat, and as drafts on LinkedIn — there, a mistaken instruction cannot put a campaign live. TikTok and Pinterest have no such default: their create tools take the initial status from the caller, so an assistant asked for an active campaign will create an active campaign. Ask for it paused, read it back, and enable it yourself. A read-only token cannot create one at all.
Our commitments
- We never sell your advertising data, and we never use it for advertising of our own.
- Your data is used only to answer the requests you or your AI assistant make — it is not used to train AI models.
- You choose which ad accounts each connection can reach, and can change that selection at any time.
- You can disconnect any platform from your connections page, which revokes Pipeboard’s access to it.
- Enterprise administrators can configure managed authentication (SSO) for their whole organisation.
Full detail in our Privacy Policy, Security and managed authentication pages.
Connect to your AI tool
Claude
Pro / Max / Team- 1.Open claude.ai/customize/connectors
- 2.Add a custom connector with the URL above
- 3.Approve the Pipeboard consent screen
See the Claude setup guide for screenshots.
Claude Code, Warp, Cursor
and any MCP clientAdd it as a Streamable HTTP server. For Claude Code:
Per-client walkthroughs: Claude Code, Warp, ChatGPT.
What to ask it
Frequently asked questions
What is the Ads MCP server?
The Pipeboard Ads MCP is a Model Context Protocol server that connects Claude, ChatGPT, Cursor or any MCP client to every ad platform you have connected to Pipeboard — Meta, Google Ads, TikTok, LinkedIn, Pinterest, Reddit, Snapchat and Google Analytics 4 — through a single URL: https://mcp.pipeboard.co/ads-mcp. It is an aggregator: each call is routed to the platform that owns the tool, using the same authentication and account scoping as the per-platform servers.
How is this different from the per-platform MCP URLs?
It is the same tools, reachable through one connection instead of one per platform. The per-platform URLs (meta-ads.mcp.pipeboard.co, google-ads.mcp.pipeboard.co, …) keep working and are the better choice when you deliberately want an assistant scoped to a single platform. Use the unified URL when you want cross-platform work — comparing spend across platforms, or one assistant that can act wherever the work is.
How are the tools named?
Every tool is namespaced by its platform, so names are unambiguous across platforms: get_campaigns becomes meta_ads_get_campaigns, and Google Ads reporting is google_ads_get_campaign_metrics. Each tool also carries a human-readable title and a read-only or write annotation, so your client can tell reads from writes before you approve a call.
There are hundreds of tools — how does the assistant find the right one?
The server exposes a tool_search tool. Ask for what you want to do ("create a Pinterest ad group") and it returns the best-matching tool definitions, ranked, so the assistant can work from a short list instead of the full inventory. You can scope a search to one platform by including its prefix, e.g. tool_search(query="meta_ads insights").
What can it reach in my ad accounts?
Only the platforms you have connected, and within those, only the ad accounts you selected when you configured access. Nothing outside that selection is reachable. A platform you have not connected returns connect guidance rather than data.
Can I give an assistant read-only access?
Yes. API tokens carry their own read-only or read-write scope, enforced server-side on every call, and read-only team members are limited the same way — a read-only credential cannot call a write tool no matter what the prompt says, so a reporting-only session cannot spend. On the write side, new campaigns are created paused by default on Meta, Google Ads, Reddit and Snapchat and as drafts on LinkedIn; TikTok and Pinterest take the initial status from the caller, so ask for those paused and enable them yourself after reading them back.
Which AI clients does it work with?
Any MCP-compatible client. It works with Claude (Pro/Max, Claude Code and the Claude apps), ChatGPT developer mode, Cursor, Warp, n8n and custom agents. Add the URL as a Streamable HTTP MCP server; Claude and other DCR-capable clients complete OAuth automatically.
How long does setup take?
Under two minutes: sign up for Pipeboard, connect at least one ad platform at pipeboard.co/connections, then add https://mcp.pipeboard.co/ads-mcp to your MCP client. Connecting more platforms later adds their tools to the same connection — you do not reconfigure the client.