MCP server · v1.0.0

Ask your AI whether to kill that ad.

Connect SellTheClick to ChatGPT, Claude, Claude Code or Cursor. Your agent gets three Meta ad tools that answer with real maths and real benchmarks.

Freeno account, no key

3 toolsall read-only

Nothing savedyour numbers are not stored

https://www.tryselltheclick.com/mcp

Quick start

Connect in one minute.

Claude (web and desktop)

Open Settings, then Connectors.
Choose Add custom connector, name it SellTheClick and paste the server URL.
In a chat, turn the SellTheClick connector on and ask a question.

ChatGPT

Open Settings, then Apps & Connectors. Under advanced settings, turn on developer mode.
Create a connector, name it SellTheClick and paste the server URL. Authentication: none.
Start a chat and pick SellTheClick from the tools menu.

Custom connectors depend on your ChatGPT plan and workspace settings.

Claude Code

claude mcp add --transport http selltheclick https://www.tryselltheclick.com/mcp

Cursor and other MCP clients

{
  "mcpServers": {
    "selltheclick": { "url": "https://www.tryselltheclick.com/mcp" }
  }
}

Try asking

"My ad spent $450 and got 6 sales. Each sale leaves me $40. Keep it or kill it?"

"Last 7 days: 44,300 impressions, 310 link clicks, 5 purchases, $310 spent. Skincare. Why isn't it selling?"

"I have 4 new ads, $50 a day, and a sale usually costs $75. How should I test them?"

Reference

Tools.

keep_or_kill

Should this Meta ad keep running? Answers keep, kill or too early.

Input	Type	Meaning
`profit_per_sale`	number, required	What one sale leaves the store before ad costs: price minus product, shipping, fees and returns (break-even CPA).
`amount_spent`	number, required	What the ad has spent so far.
`sales`	integer, required	Purchases the ad brought.

Returns headline, verdict (keep, kill or wait), next_step, cost_per_sale, net_so_far, chance_it_makes_money, more_spend_to_be_sure and a link to the full tool.

$210 down. Kill it.profit_per_sale 40, amount_spent 450, sales 6. Each sale cost $75. Chance it makes money: 7 in 100.

Method. A Poisson model asks how often luck alone would give results this good if each sale really cost exactly what it earns. Keep needs a chance of 80 in 100 or more and at least 3 sales; kill is 20 in 100 or less. These cut-offs are our choice, not a Meta rule.

find_where_ad_breaks

Why isn't this ad selling? Finds whether it breaks before the click, after it, or on price.

Input	Type	Meaning
`impressions`	number, required	Last 7 days, from Ads Manager.
`link_clicks`	number, required	Last 7 days.
`purchases`	number, required	Last 7 days.
`amount_spent`	number, required	Last 7 days.
`industry`	string, optional	For example skincare, supplements, apparel or food. Picks the right benchmarks.

Returns headline, detail, where_it_breaks, steps (click rate, buy rate and cost per sale, each with the typical figure), the benchmark source, and a link to the full report.

People see your ad and scroll past.Click rate 0.70% against a typical 2.39%. Buy rate 1.61% against 1.53%. Cost per sale $62.00 against $39.31.

Benchmarks. Triple Whale, Facebook ad benchmarks by industry, 40,000+ ecommerce brands, 1 August 2025 to 31 July 2026.

plan_ad_test

How many new ads to test, for how long, and with which setup.

Input	Type	Meaning
`new_ads`	integer, required	Number of new ads, 1 to 30.
`budget_per_day`	number, required	Daily budget for the test.
`cost_per_sale`	number, required	Usual cost per purchase (CPA).
`cost_per_1000_views`	number, optional	CPM, used when sales are too few to judge on. Default 15.

Returns headline, how_many, how_long, setup (one ad set or ABO), sales_per_week, meta_learning and a link.

Test 3. 7 days. One ad set.new_ads 3, budget_per_day 300, cost_per_sale 25.

Method. Each ad needs about 3 sales for a first read. Meta's learning-phase guideline is about 50 optimisation events in 7 days per ad set (Meta Business Help Centre).

Details

Protocol, limits and errors.

Endpoint: POST https://www.tryselltheclick.com/mcp
Transport: MCP Streamable HTTP, stateless. Every reply is a single JSON-RPC 2.0 response; no session or event stream. GET returns 405.
Protocol versions: 2025-06-18, 2025-03-26 and 2024-11-05.
Authentication: None. The tools only do maths on the numbers you send.
Tool safety: All tools are marked read-only and idempotent. None of them reads your ad account, changes anything or spends money.
Data: The numbers you send are used for the answer and are not saved to a database. See the privacy policy.
Rate limits: Fair use. Automated bulk calls may be slowed down.
Cost: Free. The full report on the website adds competitor comparisons, buyer questions and new angles.

Errors

A bad or missing input returns a normal tool result with isError: true and a message that names the inputs needed, so the agent can fix the call and try again. An unknown method returns JSON-RPC error -32601; a body that is not JSON returns -32700.

{"content": [{"type": "text", "text": "Could not run keep_or_kill: profit_per_sale and amount_spent must be above 0, and sales 0 or more. It needs: profit_per_sale, amount_spent, sales."}], "isError": true}

Changelog

Versions.

1.0.0 · 13 November 2026. First release: keep_or_kill, find_where_ad_breaks, plan_ad_test.

Support

Questions or a bug?

Email support@tryselltheclick.com with the tool name, the inputs you sent and what came back.

Run a full report free