Skip to content

Repository files navigation

WorkMagic for AI assistants

WorkMagic for AI assistants — multi-touch attribution, incrementality and MMM in Claude, ChatGPT, Codex, Cursor, Copilot and Gemini

Ask your marketing measurement data anything — in Claude, ChatGPT, Codex, Cursor, Copilot or Gemini. Attribution by channel and campaign, why a number moved, lift test results and media budget plans, answered from your own WorkMagic account.

Add to Claude ChatGPT Codex Cursor Gemini CLI MCP License: MIT Follow on LinkedIn

⭐ Star this repo to follow new skills and platforms as they ship — see the changelog.

What is WorkMagic

WorkMagic is an AI-powered marketing measurement platform for ecommerce and DTC brands. It brings multi-touch attribution (MTA), incrementality (geo lift tests) and media mix modeling (MMM) / budget planning together — across Meta, Google, TikTok and more — so you can see what's actually driving sales and decide where to spend next.

This open-source (MIT) plugin lets your AI assistant ask that same data questions in plain language, using your own WorkMagic account — an open-source MCP server plus skills for marketing attribution, incrementality and MMM.

See it in action

You: Compare Meta and Google attributed revenue under data-driven, last click, first click and any-click attribution for the last 30 days, and explain why they differ.

Claude: (abridged — WorkMagic demo store, synthetic data)

Channel Data-driven Last click First click Any click
Meta $10.2K $17.3K $27.1K $38.0K
Google $12.7K $19.9K — $33.2K

Meta looks strongest under first- and any-click because it starts most journeys: shoppers first arrive from a Meta ad, then come back through Google or direct before buying. Data-driven attribution splits each order's credit across those touchpoints, so both channels shrink. Any-click gives every touched channel full credit, which is why the two any-click figures add up to more than the store actually sold ($46.9K).

More sample conversations: readiness check · model comparison · best-selling products

What's included

The plugin has two parts: the WorkMagic MCP server (https://mcp.workmagic.io/mcp), which reads and updates your WorkMagic account after you sign in, and nine skills that tell the assistant how to use it well.

Skill Use it when you want to
attribution-data-query Get attribution numbers — ROAS, CAC, revenue, orders by channel, campaign or period
attribution-intent-clarification Ask a vague question; the assistant asks one focused follow-up, then answers
attribution-model-comparison See iDDA, DDA, last-click and platform-reported numbers side by side, and why they differ
attribution-anomaly-diagnosis Understand why attribution dropped, spiked or shows zero
attribution-custom-dimension Break results down by audience, region, brand or another label in your campaign names
attribution-edge-routing Get pointed to the right place when a question is outside attribution
lift-test-readout Interpret a completed lift test (incremental ROAS, confidence) and decide what to do next
mbo-read-scenario Read, compare or reconcile a Media Budget Optimizer scenario
mbo-create-scenario Ask "how should I split my budget?" and get a configured MBO scenario

Try asking:

  • "What was my Meta ROAS last week?"
  • "Why did Google attribution drop on Tuesday?"
  • "Which products sold the most in the last 30 days?"
  • "How should I split a $200K budget next month?"
  • "Is my account ready to run a Meta lift test?"

Get WorkMagic

Using this plugin needs an active WorkMagic account.

  • New to WorkMagic? See plans and pricing to get started.
  • Already a customer? Install below and sign in with your WorkMagic account the first time the assistant uses it.

Questions about plans? Contact your WorkMagic CSM or support@workmagic.io. Lift test and budget optimizer skills need those features enabled on your account.

Install

Claude (claude.ai, desktop, Cowork)

  1. Open plugin settings (Settings → Plugins, or Plugins & skills), choose Add marketplace → GitHub repo, enter workmagicio/workmagic-agent-plugin, then install WorkMagic. (Listing in the Claude plugin directory is coming soon.)
  2. Open the plugin's Connectors tab and connect WorkMagic.

Only want the data connection? Add https://mcp.workmagic.io/mcp under Customize > Connectors as a custom connector.

Claude Code

/plugin marketplace add workmagicio/workmagic-agent-plugin
/plugin install workmagic@workmagic-plugins

Codex

codex plugin marketplace add workmagicio/workmagic-agent-plugin

Then install workmagic from /plugins.

Gemini CLI

gemini extensions install https://github.com/workmagicio/workmagic-agent-plugin

ChatGPT, Cursor, GitHub Copilot

Directory listings are coming soon. Until then, connect the MCP server below; in Cursor you can also add this repository as a team marketplace.

Any other MCP client

Add https://mcp.workmagic.io/mcp as a remote (Streamable HTTP) MCP server. You get the tools; the skills in skills/ are plain Markdown you can load as instructions.

The first time the assistant uses WorkMagic, sign in with your WorkMagic account. If you belong to several WorkMagic accounts, the assistant asks which one to use. Don't have an account yet? See Get WorkMagic above.

Data

The skills are instructions only; they don't run code or contact any other service. The assistant reaches your data through the WorkMagic MCP server at https://mcp.workmagic.io/mcp, using the account you sign in with. The server receives the requests the assistant makes on your behalf, such as queries, filters and settings you ask to change, and returns your WorkMagic data. WorkMagic logs those requests for security and support. Nothing is sent anywhere else.

Some skills can change settings in your WorkMagic account, such as creating or editing a budget scenario. The assistant shows what it will change and asks you to confirm first.

The tools/ folder holds scripts WorkMagic uses to maintain this repository. The plugin never runs them.

Follow WorkMagic

New skills, platforms and product updates ship often — follow along:

Support

Contact your WorkMagic CSM or support@workmagic.io. Privacy policy: www.workmagic.io/privacy. Found a problem with a skill? Open an issue.

Releases

Packages

Contributors

Languages