Talk to your campaigns. Claude answers with real data.
Paid Media MCP is an open-source server that connects Claude to your Google Ads, Meta Ads and TikTok Ads accounts with read-only access. Ask in plain language; get analysis on real data. No dashboards, no exporting CSVs, nothing sent outside your machine.
An open protocol between your platforms and the AI
MCP (Model Context Protocol) is Anthropic's open standard for connecting AI assistants to external data. The server exposes your accounts as tools that Claude can query, all normalized into a single format.
Google Ads
Meta Ads
TikTok AdsClaude Desktop · Claude Code · any MCP client
Ask the way you'd talk to your analyst
Four tool areas, all read-only and normalized across Google, Meta and TikTok. Claude picks the right tool based on what you ask.
Performance 6 tools
Spend, conversions, ROAS by campaign and by day. Normalized across all 3 platforms.
Temporal 2 tools
Hour of day × day of week. Find out when your spend actually works.
Diagnostics 6 tools
Search terms, Quality Score, impression share, z-score anomalies, auction insights.
Normalized 2 tools
CTR, CPC, CPM, CPA, conversion value and ROAS with the same name everywhere.
visibility Read-only: Claude reads, it never touches campaigns or bids.
And this is only the beginning
The free MCP covers paid media. But the same approach scales to your whole business: I connect Claude to your CRM, ERP and CMS too and build a custom dashboard where spend meets real pipeline, margins and content.
- hub CRM → leads, SQLs and won deals next to spend
- payments ERP → real margin, not just platform ROAS
- article CMS → which content feeds which campaign
- dashboard Ad-hoc dashboard, built for your operation
Up and running in 4 steps
You need Node ≥ 20 and your platform tokens. Full documentation is in the repo.
# 1 · Clone the repo
$ git clone https://github.com/Pauesome/Paid-Media-MCP
# 2 · Install and build
$ npm install && npm run build
# 3 · Add your credentials
$ cp .env.example .env
# 4 · Connect to Claude (stdio), done ✦
From export-to-Excel to a dialogue with your data
The usual flow
- closeOpen 3 platforms, export CSVs, paste them into a sheet
- closeMetrics with different names in each panel
- closeAnomalies show up once you've already burned the budget
- closeUploading your data to a third-party SaaS tool
Ask and receive
- check"Where do I move spend?" in plain language
- checkMetrics normalized across Google, Meta and TikTok
- checkAnomaly detection by z-score, not by eye
- checkLocal credentials, zero data sent outside
Prefer I set it up for you?
The MCP is free and open source: install it yourself. But if you want it left connected to your stack, with custom skills and subagents (audits, budget strategy, a dashboard crossing paid with CRM and ERP), I'll build it. Tell me your situation and I'll reply within 24h.
- check_circle Full setup and connection to your accounts
- check_circle Custom skills and subagents on top of the MCP
- check_circle Ad-hoc dashboard with CRM · ERP · CMS