AdWhispr MCP Server
Chat with any brand's Meta (Facebook/Instagram) ads inside Claude — research a competitor's ad library, find their longest-running winners, extract hooks and formats, and clone winning ads for your own brand.
README
adwhispr-mcp-server
Chat with any brand's Meta ads inside Claude. This connects Claude to AdWhispr — research a competitor's Facebook/Instagram ad library, find their longest-running winners, and clone the creative for your own brand, all from a Claude conversation.
It's a thin, open-source bridge to the AdWhispr remote MCP server (https://adwhispr.com/api/mcp). Sign-in happens automatically through your browser the first time.
Install (Claude Desktop) — 3 steps
npx adwhispr-mcp-server config
- Run the command above. It adds AdWhispr to your Claude Desktop config.
- Fully quit and reopen Claude Desktop.
- Start a chat — a browser window opens once to sign in to AdWhispr. Done.
Then try:
"Search AdWhispr for Nike's longest-running ads." "What hooks is Hims using in their Meta ads?" "Clone Cal AI's top ad for my brand at example.com."
Manual setup
Prefer to edit the config yourself? Add this to claude_desktop_config.json:
{
"mcpServers": {
"adwhispr": {
"command": "npx",
"args": ["-y", "adwhispr-mcp-server", "serve"]
}
}
}
Config file location:
- macOS
~/Library/Application Support/Claude/claude_desktop_config.json - Windows
%APPDATA%\Claude\claude_desktop_config.json - Linux
~/.config/Claude/claude_desktop_config.json
Other MCP clients
Any MCP client that runs a stdio command works — point it at npx -y adwhispr-mcp-server serve, or connect directly to the remote endpoint https://adwhispr.com/api/mcp if your client supports remote MCP servers with OAuth.
Tools
| Tool | What it does |
|---|---|
search_brands |
Find brands tracked in AdWhispr by name (returns ad counts + IDs). |
get_brand_ads |
Get a brand's ads with hooks, formats, performance signals, and creative links. |
get_brand_stats |
Aggregated stats for a brand: ad counts, hook/format distributions, spend ranges. |
search_ads |
Semantic search across a brand's ads by concept ("before/after", "social proof"). |
compare_brands |
Compare 2–3 brands side by side. |
add_brand |
Start tracking a new competitor brand (kicks off Meta ingestion). |
clone_ad |
Clone a competitor's winning ad for your brand — a real generated image for image ads, a scene-by-scene script for video ads. |
Pricing
- Free — 5 tool calls/month, 1 tracked brand, 1 free clone.
- Pro — $29/mo (3-day free trial): unlimited tool calls, 3 brands, 10 clones/month.
- Agency — $149/mo: 10+ brands, 50 clones/month, cross-brand comparison.
Start a trial at adwhispr.com/upgrade.
How it works
Claude Desktop launches npx -y adwhispr-mcp-server serve, which runs mcp-remote pointed at https://adwhispr.com/api/mcp. That bridges Claude's local stdio transport to the AdWhispr server over HTTP and handles the OAuth login. Your ad data and account live on AdWhispr; this package stores nothing.
To point at a different server (self-hosting/testing), set ADWHISPR_MCP_URL.
Troubleshooting
AdWhispr doesn't appear in Claude after install.
Fully quit Claude Desktop (not just close the window) and reopen it — config is only read on startup. Confirm the entry exists in claude_desktop_config.json (paths above). On macOS/Linux make sure npx is on your PATH; if Claude can't find it, set the command to the absolute path from which npx.
The browser sign-in window never opens, or login loops.
The first tool call opens a browser to authorize via OAuth. If it doesn't appear, check that your default browser can open and that no firewall is blocking localhost callbacks. Clearing the stale auth cache fixes most loops:
rm -rf ~/.mcp-auth
Then restart Claude Desktop and trigger a tool call again.
"Authentication required" or tools return an auth error.
Your AdWhispr session expired or wasn't completed. Re-run the sign-in (clear ~/.mcp-auth as above), or sign in directly at adwhispr.com first, then retry.
A tool returns an upgrade / out-of-quota / "locked clone" message. That's expected on the Free tier (5 tool calls/month, 1 brand, 1 clone). The message includes an unlock link — open it to upgrade or buy a credit pack. Relay the link as-is; the clone is generated and waiting behind it.
add_brand says a brand isn't found, or search_brands returns nothing.
add_brand resolves brands from the Meta Ad Library by name — try the exact brand name as it appears on Facebook. After adding, ingestion runs in the background (~40s) before ads are queryable. search_brands only returns brands already tracked on your account.
Node / npx errors on launch.
Use Node 18+ (node --version). If an old cached package is misbehaving, force a fresh copy: npx -y adwhispr-mcp-server@latest serve.
Connecting a non-Claude-Desktop client.
Point any stdio MCP client at npx -y adwhispr-mcp-server serve, or connect directly to the remote endpoint https://adwhispr.com/api/mcp if your client supports remote MCP servers with OAuth.
Still stuck? Open an issue at github.com/adwhispr/mcp-server/issues or email basil@adwhispr.com.
Links
- Website: https://adwhispr.com/connect?ref=mcp_directory
- Tutorial: How to chat with any brand's Meta ads inside Claude in 90 seconds — https://adwhispr.com/blog (coming soon)
- Issues: https://github.com/adwhispr/mcp-server/issues
License
MIT © AdWhispr
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。