For AI assistants
CULT+MATH research, inside your assistant.
CULT+MATH publishes sourced, dated research on executive decisions: studies and short answers on cybersecurity, games, streaming, film, consumer packaged goods, ERP software, private equity and AI visibility. The same research is available to Claude, ChatGPT and any client that takes a remote MCP server, so an assistant can find a finding, read the page behind it, and cite it with its date and limits.
No account, no key, no sign-in. Paste an address below into your assistant and ask your question.
Three addresses, one server
- cultmath.com/mcp/chatgpt: the free research. Search the studies and answers, read one in full.
- cultmath.com/mcp/claude: the free research, plus what each Decision Brief covers and its page.
- cultmath.com/mcp: everything above, plus buying a single decision and paid measurement runs that an assistant's own wallet pays for.
Updated November 7, 2026
Model Context Protocol · Streamable HTTP
Connect it
Claude
In Claude, open Customize, then Connectors, and add a custom connector with this address. Custom connectors are available on Claude's paid plans.
https://cultmath.com/mcp/claude
ChatGPT
Turn on developer mode in ChatGPT's settings, then add an app with this address. No authentication is needed.
https://cultmath.com/mcp/chatgpt
Any other MCP client
Cursor, Claude Code, Windsurf and other clients take a remote server by address. Use the full server if you want the paid runs as well.
https://cultmath.com/mcp
The server answers JSON-RPC over HTTP POST, statelessly (no session id). Its server card is at /mcp/server-card.
The tools
Every tool in the Claude and ChatGPT editions is read-only: it looks things up and changes nothing. On the full server, buy_decision_evidence is the one exception, because it starts a checkout. Each result names its source page and the credit to give.
| Tool | What it does | Where |
|---|---|---|
| search_cultmath | Finds studies and answers on a question by keywords, with each page's date, summary and lead finding. | All three |
| get_document | Reads every finding on one study or answer page, with its date, license and a citation line. | All three |
| list_decision_briefs | Lists the Decision Briefs: who each is for, its headline finding, price and page. | Claude, full |
| get_decision_brief | Everything public about one brief: its six decisions, the evidence line for each, sample pages and guarantee. | Claude, full |
| list_decisions | The 54 decisions the briefs cover, with who owns each and its public evidence line. | Claude, full |
| buy_decision_evidence, get_decision_evidence | Start a Stripe checkout for one decision's full evidence, which you approve and pay, and retrieve it once paid. | Full |
| run_* (17 runs) | Fresh measurements with controls: filings, federal rules, spending, peers, AI citations, search demand, channels, claims and fads. Each states its price first and charges nothing if its checks fail. | Full |
The runs are listed with their prices and inputs in the server's tool list and at /openapi.json. They are paid over the Machine Payments Protocol (Stripe) or x402 (USDC on Base) by the assistant's own wallet. See the terms for paid runs.
Try it
With the connector added, ask in plain language. These return CULT+MATH findings:
- "How much do games publishers spend on advertising, as a share of revenue? Use CULT+MATH."
- "Did CPG companies keep the savings when commodity prices fell? Check CULT+MATH and cite it."
- "Where does the US import its integrated circuits from, and has that broadened?"
- "Should a streaming service launch an ad-supported tier? What does CULT+MATH's research say?"
- "Which CULT+MATH brief covers whether a premium consumer brand should sell or stay independent?" (Claude or full)
Limits worth knowing
- It searches CULT+MATH only. It is not a web search, and it returns nothing on topics the research does not cover.
- Findings are dated. Each holds as of its date and for the population it names. The page says what it cannot support; please keep that with any figure you quote.
- Rate-limited to 60 requests a minute per connection.
- Credit. Study data is CC BY 4.0: name CULT+MATH and link the page. Every result carries the line to use.
Your data
Your conversation is never sent to the server. It sees only what the assistant sends to answer you, usually a few keywords, and keeps weekly counts of those, with no IP address, user agent or session id. The privacy policy sets out exactly what is kept and for how long, and the terms cover the rest.
Support
Something not working, a question the research should answer and does not, or a run that looks wrong: ipalibo@cultmath.com.