> For the complete documentation index, see [llms.txt](https://q3labs.gitbook.io/polypay-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://q3labs.gitbook.io/polypay-docs/llms-txt-for-agents.md).

# PolyPay for AI Agents

PolyPay ships an [llms.txt](https://llmstxt.org) playbook that lets any AI agent — Claude, ChatGPT, Cursor, autonomous trading bots, custom LangChain workflows — read, understand, and operate PolyPay through plain HTTP without you having to write a custom integration.

## What is llms.txt?

[`llms.txt`](https://llmstxt.org) is a simple convention: a single markdown file at a well-known URL that tells LLMs how to interact with a service. Think of it as a robots.txt, but for AI agents instead of crawlers. The agent fetches the file, reads the instructions, and can then call the right endpoints with the right payloads on its own.

## Live endpoint

```
https://api.polypay.pro/llms.txt
```

Plain text, CORS-enabled, cached for 1 hour. No authentication needed to read.

## What's inside?

A complete playbook covering the five most common agent flows:

| # | Flow                            | What the agent can do                                                                                                         |
| - | ------------------------------- | ----------------------------------------------------------------------------------------------------------------------------- |
| 1 | **Login**                       | Generate a ZK auth proof and obtain a JWT                                                                                     |
| 2 | **Create multisig account**     | Deploy a new PolyPay multisig on Horizen, Base, or Arbitrum (see [Arbitrum Stylus Support](/polypay-docs/arbitrum-stylus.md)) |
| 3 | **Single transfer**             | Propose, vote, and execute a private payroll payment                                                                          |
| 4 | **Batch transfer**              | Propose and execute multi-recipient payroll in one transaction                                                                |
| 5 | **Gasless USDC deposit (x402)** | Fund any PolyPay multisig with USDC on Base without holding ETH                                                               |

Each flow includes the exact HTTP requests, payload shapes, ZK proof generation snippets (Noir + UltraHonk), and links to the compiled circuit artifacts the agent needs.

The full live OpenAPI / Swagger spec sits next to it at `https://api.polypay.pro/swagger` for endpoints not covered in the playbook.

## How to point an agent at PolyPay

Most agent frameworks support custom system prompts or tools. Two common ways to wire up PolyPay:

**1. Drop the URL into your prompt.**

> Fetch `https://api.polypay.pro/llms.txt` and use it to send 100 USDC from my PolyPay multisig `0x...` to the addresses in `recipients.csv`. Sign all proofs with the wallet I'm connected to.

The agent will read the playbook and follow the steps end-to-end.

**2. Add it as a tool / resource.**

If you're building with Claude Code, OpenAI Custom GPTs, Cursor, or any MCP-compatible agent, register `https://api.polypay.pro/llms.txt` as a documentation source. The agent gets persistent context for every session.

## What can agents actually do?

A few non-trivial examples that work out of the box:

* *"Run this week's payroll on PolyPay. Recipients and amounts are in the attached CSV. Use my Base mainnet multisig."*
* *"Top up the team's PolyPay treasury with 5,000 USDC from my Base wallet, gasless."*
* *"Show me the last 30 days of payroll history for our PolyPay multisig, grouped by recipient."*
* *"Propose a transfer of 1 ETH on Horizen to the address from the Telegram message, wait for the other signers, then execute when the threshold is met."*

## Compatibility

Tested against agents using:

* Anthropic Claude (Claude Code, claude.ai with browsing)
* OpenAI ChatGPT (with browsing or Custom GPT)
* Cursor / Copilot agents that follow `llms.txt`
* LangChain / LangGraph custom agents
* MCP servers exposing the URL as a resource

If your framework can fetch a URL and follow markdown instructions, it can drive PolyPay.

## For developers

The playbook is generated server-side from [`packages/backend/src/llms-txt/llms-txt.content.ts`](https://github.com/Poly-pay/polypay_app/blob/develop/packages/backend/src/llms-txt/llms-txt.content.ts). Updates to the content go through the normal PR flow — the live endpoint reflects the latest deployed commit on `main`.

Open an issue on [GitHub](https://github.com/Poly-pay/polypay_app/issues) if you spot an out-of-date example, a missing flow, or an agent framework that can't parse the playbook cleanly.