Build with AI agents

An AI coding agent can integrate Fluveo by reading these docs and nothing else. This page lists what we publish for agents, what a person has to do first, and the limits an agent meets.

What a person does, and what the agent does#

A person does two things the agent cannot do:

Create the Fluveo account in the Dashboard. Test mode works at once; activate the account in the Dashboard only to go live.
Create a test secret key (sk_test_...) in the Dashboard under Settings > API keys, with the Create secret key button and a name for the agent, and put it in an environment variable the agent's code reads, such as FLUVEO_SECRET_KEY. The Dashboard shows a new secret key only once. An agent that only reads can get a restricted key (rk_test_...) with Read on the resources it needs instead. For a checkout page with Fluveo's payment form, also copy the test publishable key (pk_test_...) from the same screen into FLUVEO_PUBLISHABLE_KEY: it is the only key a page may hold.

The agent does the rest in test mode: it writes the integration, calls the API, creates the webhook endpoint, verifies signatures and runs the checks at the end of the integration playbook.

What we publish for agents#

Resource	Where	What it is for
llms.txt	/llms.txt	The index: what Fluveo is, the hard rules, and one line per page. Start here.
One file per task: the checkout	/llms/custom-checkout.txt	Everything to build a first checkout page of your own with Fluveo's payment form, under 40 KB: the rules, the quickstart, the calls a checkout makes and the errors they throw, the test cards and the webhook signature check.
One file per task: the payment form	/llms/payment-element.txt	What changes the form after a first checkout, under 40 KB: the rules, the form's calls, events and options, and the errors they throw.
Every page in one file	/llms-full.txt	All pages, each starting with its title and a `Source:` line. For agents with a long context.
Markdown of any page	the page address plus `.md`, such as /agents/integrate.md	The page as clean Markdown: every code tab labelled, and the notes written only for agents.
Markdown in one language	`.curl.md`, `.node.md`, `.python.md` or `.php.md` in place of `.md`, such as `/agents/integrate.python.md`	The same Markdown with each code group using that language when present, otherwise cURL when present or its first tab. Much shorter on long pages. Each `.md` lists its copies at the top.
OpenAPI	/openapi.json	Every operation Fluveo serves. If it is not in this file, it does not exist.
API version	/versions.json	The one API version Fluveo serves, `2026-05-27.dahlia`.
Agent skill	/skills/fluveo-custom-checkout/SKILL.md	A skill file for coding agents that load skills: the rules, the steps and the checks of a custom checkout. See Install the skill.
MCP server	The Fluveo MCP server is not available in production yet.	For agents that connect to servers instead of fetching pages. See Fluveo MCP.

Every HTML page also links its Markdown copy with <link rel="alternate" type="text/markdown">, so a tool that reads the HTML can find it.

Next to the title of every page, the Use with AI menu has four entries:

Copy for LLM copies the page's Markdown with Fluveo's hard rules on top, ready to paste into any chat.
View as Markdown opens the .md copy.
Open in ChatGPT and Open in Claude start a chat that reads the page.

Ready prompts#

The ready prompts are two blocks of text a person pastes into their coding agent, each with the rules the agent must follow and the checks it must run before it says it is done:

"Build a custom checkout page for this app with Fluveo": your own page, with Fluveo's payment form in it.
"Integrate Fluveo payments into this app": payments made from your server, with no page of ours.

Install the skill#

A coding agent that loads skills can keep the custom checkout rules at hand. Download SKILL.md and save it in a folder named fluveo-custom-checkout inside the folder your agent reads skills from. For an agent that reads .agents/skills in the project, the file is .agents/skills/fluveo-custom-checkout/SKILL.md.

The skill is short on purpose: the rules, the steps and the checks. It sends the agent to /llms/custom-checkout.txt for the code, and to /llms/payment-element.txt for the form's options.

Connect the MCP server#

The Fluveo MCP server is not available in production yet.

It lists 28 tools, each shown only to a key whose permissions allow it. search_fluveo_docs and fetch_fluveo_doc read these docs. fluveo_api_search, fluveo_api_details, fluveo_api_read and fluveo_api_write find and run API operations. get_account_info names the store behind the key. Seven more each read one kind of store data: get_balance, list_disputes, retrieve_dispute, list_events, retrieve_event, list_webhook_endpoints and retrieve_webhook_endpoint. list_customers, retrieve_customer, create_customer and update_customer read and change the store's customers. create_checkout_session and create_payment_link make a Fluveo payment page or a payment link for a buyer. list_payments, retrieve_payment, search_payments, list_refunds and retrieve_refund read the store's payments and refunds. create_refund, capture_payment and cancel_payment refund, capture or cancel a payment, each only after a person approves that one call in the Dashboard. MCP tools has each one's arguments and answers.

Limits#

Markdown has its own address. A page always answers HTML, even to Accept: text/markdown. Ask for the .md address instead.
The MCP server starts no payment. Through it an agent reads the docs and the store, and runs the writes its agent key allows, with a person approving the sensitive ones in the Dashboard. An agent key may not create or confirm a PaymentIntent, or create a payout. An agent's code creates payments with a secret key and plain HTTP calls to the API.

Next#

Integrate Fluveo with an AI agent: the step-by-step playbook.
Fluveo MCP: let an AI client read and change your test store through the MCP server.
The ready prompts: copy one into your coding agent.
Quickstart: your first payment, written for people.
Testing: test keys and test values.