marketing-lab-mcp

Marketing math your AI can call

Nine marketing tools (A/B tests, budget splits, attribution, customer segments) that Claude calls to run the numbers instead of guessing them.

Server URL
Live URL coming soon

The public server is not deployed yet. Run it on your machine in the meantime.

A recorded conversation: Claude picks the tools, then answers

    How it works

    One createServer() call builds the MCP server and its nine tools. Three runtimes wrap that same server in a different transport, so a tool gives the same answer wherever it runs.

    1. In your browser

      The playground on this page imports createServer() and connects to it over an in-memory transport. The call never leaves the tab. 

      {"jsonrpc": "2.0", "id": 2,
 "method": "tools/call",
 "params": {
  "name": "ab_test_plan",
  "arguments": {
   "baseline_rate": 0.03,
   "mde_relative": 0.2,
   "daily_visitors": 2000}}}

    2. On a Cloudflare Worker

      The Worker builds a fresh server for every request and answers Streamable HTTP at /mcp. That URL is what you paste into Claude as a custom connector. 

      {"jsonrpc": "2.0", "id": 2,
 "method": "tools/call",
 "params": {
  "name": "budget_optimize",
  "arguments": {}}}

    3. On your machine

      npm run start:stdio runs the same server over stdin and stdout, one JSON-RPC message per line, so Claude Desktop can start it as a local command. 

      {"jsonrpc": "2.0", "id": 2,
 "method": "tools/call",
 "params": {
  "name": "store_sales",
  "arguments": {
   "group_by": "category"}}}

    Transcripts

    Loading four recorded conversations with Claude.

    The nine tools

    Each tile runs a tool's first example in your browser, with the same code the server uses. Pick a tool name to open it in the playground.

    • A/B tests

      Did the variant really win? And how many visitors does the next test need?

      Running the example in your browser.

      ab_test_evaluate, example "Hillstrom-sized email test"

    • Attribution

      Five attribution models side by side, scored on a simulated sample whose true shares are known.

      Running the example in your browser.

      attribution_compare, example "Simulated sample with known truth"

    • E-commerce

      Sales by category, state or month, RFM customer segments and a B2B lead funnel, from Olist's public data.

      Running the example in your browser.

      store_sales, example "Top 10 categories by revenue", top 5 shown

    • Budget

      What a weekly media budget buys, and the split a marketing mix model expects to sell more.

      Running the example in your browser.

      budget_simulate, example "Current plan", against budget_optimize, example "Reallocate the current budget"

    Playground

    Loading the nine tools. They run in your browser, with no sign-up.

    Connect to your Claude