Tell your agent to set it up.
It installs and runs the whole program.
Claude Code, Cursor, Codex, or any agent that can fetch a URL reads Qanary's
llms.txt, wires the tracking snippet and checkout, and reports back what
it verified — no dashboard clicking required.
Once you're live, the same chat window keeps running the program through Qanary's MCP server: campaigns, payouts, attribution — all in plain English.
- One prompt installs the snippet and stamps checkout
- MCP server for Claude Code, Cursor, Codex & more
- No plugin, no SDK — just a URL any agent can fetch
Two prompts. That's the whole surface.
The first prompt is the entire install — no manual snippet placement, no OAuth dance to click through by hand. The second is how you run things afterward: reward tweaks, roster questions, payout checks, all from the same chat window.
Read https://qanary.cloud/llms.txt and install Qanary on my site:
add the snippet, wire my checkout, and tell me what to verify.
Works with Claude Code, Codex, Cursor — whatever you vibe with.
Your agent reads llms.txt, installs the snippet, wires the
checkout, and reports what it verified.
Spin up a Black Friday campaign at 25% for my top affiliates,
then show me last month's payouts.
Point the same agent at Qanary's MCP server and it runs
the program in plain English — approve an affiliate, pull this month's usage, or fire
diagnose_attribution to trace a stalled conversion from first click to ledger
line. The chat window becomes the control panel for your whole affiliate stack.
your agent installs it, runs it, and reports back. you read the revenue.
Bring your own agent.
No plugin to install and nothing to vendor-lock — just a file any capable coding agent can fetch and follow, plus a standard MCP endpoint for the ones that speak it.
Anything missing? If it can fetch a URL and follow instructions,
it can run llms.txt end to end.
Why hand this to an agent at all?
Idempotent by design
Every install step is safe to re-run. If your agent gets interrupted mid-install or you ask it to double-check its work, running the same prompt again never double-installs or corrupts anything.
Verifies before it declares success
The install runbook ends with a checklist, not a claim: confirm the cookie
lands, confirm checkoutStamp() returns a ref, run a test conversion.
Your agent reports exactly what it observed — and what it couldn't check from
its own environment.
No plugin, no SDK
Nothing to install into your agent, no npm package to vendor, no proprietary
protocol to learn. llms.txt is plain text any agent can fetch;
the MCP server speaks the open Model Context Protocol.
What is llms.txt?
llms.txt is a plain-text runbook written for an AI agent
instead of a human — exact steps, real code samples, and the URLs an agent needs to act, with
none of the marketing copy a person would want. Qanary's copy covers the full install, how to
stamp a checkout on each supported biller, and how to verify the result before declaring
success. Any agent that can fetch a URL and follow instructions can run it end to end.
What your agent can do via MCP
Qanary's MCP server lives at app.qanary.cloud/mcp. Point an MCP-aware client at
it — with your own account's credentials — and it can:
- Create programs — spin up a new affiliate or referral campaign, reward rules included, without opening the dashboard.
- List promoters — see who's enrolled, who's pending approval, and who's tagged for a VIP rate.
- Pull usage & plan fit — ask how close the account is to its monthly event cap before it becomes a surprise conversation.
- Run
diagnose_attribution— trace one stalled conversion from the first click to the ledger line, in plain language, when a promoter says a referral never got credited.
Switching is one more prompt, too.
Export your roster from your current platform, hand the file to your agent, and point it
at the same llms.txt — existing ?via= links
and coupon codes keep attributing after the move, so nothing your affiliates already shared
goes dark. See the full walkthrough in the switching section on the homepage.
Fair questions.
What is llms.txt?
A plain-text runbook at /llms.txt written for an AI
agent instead of a human — exact steps, real code samples, and the URLs an agent needs
to install Qanary, stamp a checkout, and verify the result. No marketing copy, nothing
to parse out.
Which AI agents does Qanary work with?
Any agent that can fetch a URL and follow instructions can run the install end to end — Claude Code, Cursor, Codex, and VS Code's agent tooling are the ones we test against most, and anything MCP-aware can operate the program afterward. There's no allowlist and nothing to request access to.
Do I need to install a plugin or SDK?
No. llms.txt is plain text, and the MCP server at
app.qanary.cloud/mcp speaks the open Model Context Protocol — nothing
proprietary to vendor into your agent or your codebase.
Is it safe to re-run the install prompt?
Yes — every step in llms.txt is idempotent. If your agent gets
interrupted, or you just want it to double-check its own work, running the same
prompt again won't double-install the snippet or duplicate anything.
What can my agent do after install, through MCP?
Create programs, list promoters, pull usage against the plan limit, and run
diagnose_attribution to trace a stalled conversion from first click to
ledger line — all in plain English, from the same chat window it installed from.
Can an agent switch me over from another platform?
Yes — export your roster from your current tool, hand the file to your agent, and
point it at the same llms.txt. Existing ?via= links and
coupon codes keep attributing after the move.
Stop clicking through your own install.
Hand your agent the file. Read the revenue when it's done.
Join the waitlist