Skip to main content
Use an AI coding agent to build an OpenFX v3 integration in the language and framework you already use. You give the agent three things: the OpenFX documentation, a short rules file that prevents the most common integration mistakes, and a prompt for the task.

Choose your starting point

The OpenFX MCP connector reads your account data and is read-only. It does not read this documentation and can’t place trades or move funds. Your coding agent writes code that calls the API with your Sandbox API key.

Before you begin

  • A Sandbox API key file. The key JSON contains kid and privateKey, and Sandbox keys carry a sandbox_ prefix. See Authentication and Environments.
  • A coding agent that reads an AGENTS.md file or a project instruction file, such as Claude Code, Codex, Cursor, or GitHub Copilot.
Build and test against Sandbox first. Sandbox and Live use the same endpoints and code, so going live means changing the API key, the webhook signing secret, and the X-App-Mode value in configuration. Never paste an API key, private key, or webhook signing secret into a chat. Give the agent the path to the key file or the name of an environment variable.

Machine-readable docs

Point your agent at these files so it reads the documentation instead of guessing. Where a guide page and the OpenAPI specification disagree, the specification is the contract.

Add the rules file and skill

Add AGENTS.md to your repository so every coding agent starts with the OpenFX v3 rules. Add the skill if you use Claude Code. Commit both so every contributor and agent session uses the same rules. They link to the documentation instead of copying it, so they stay current when the documentation changes.
AGENTS.md is a plain Markdown file at the root of your repository that coding agents read before they work. Copy this file into your repository. It states each rule briefly and links to the page that owns the detail.
AGENTS.md
Claude Code reads CLAUDE.md, not AGENTS.md. If your repository doesn’t already have a CLAUDE.md, create one that contains the single line @AGENTS.md so Claude Code loads the same rules.
A skill packages a repeatable workflow that Claude Code loads when a request matches its description. Save this file as .claude/skills/openfx-integration/SKILL.md in your repository. It sends the agent to the rules file and the matching task prompt, so the rules stay in one place.
SKILL.md
Codex, Cursor, and GitHub Copilot read AGENTS.md or their own instruction files directly. Skills are specific to Claude Code.

Prompt library

Each prompt tells the agent which pages to read, what to build, and what to deliver. The prompts point to the documentation for contract details, so they stay correct when the documentation changes.

Use a prompt

1

Copy a prompt

Choose the prompt for your task below and copy its code block.
2

Paste it into your agent

The agent asks for your language, HTTP client, and the path to your Sandbox key file. Answer with paths and names, not secrets.
3

Review and keep the output

Read the code the agent writes, run its tests against Sandbox, and commit the helpers and tests so you can rerun them.
Each prompt assumes the agent can read https://docs.openfx.com. If it can’t fetch URLs, paste the pages it names into the conversation.

Start here

Use this in a chat assistant when you don’t have a coding agent open yet.
Builds a signing helper, a thin client, and a smoke script that completes one Sandbox trade.

Build a flow

Builds the conversion flow: discover pairs, quote, execute before expiry, and track the trade.
Builds settle and reverse flows for the settlement model your organization uses.
Builds recipients, payout addresses, quotes, payouts, confirmation, and lifecycle tracking.
Builds one receiver for every event family with signature verification, deduplication, and state tracking.
Builds persisted idempotency keys, crash recovery, and error-driven retries for every write.

Debug and review

Builds a tested JWT and X-Request-Signature helper, or diagnoses 401 and 403 responses from an existing one.
Audits an integration against the v3 contract and ranks findings by risk to money.
Verifies readiness and produces a go/no-go report before you switch to Live.

Verify what the agent built

An agent can produce code that looks correct and still fails on the first Live request. Before you rely on it:
  1. Run the integration against Sandbox and confirm it completes a trade or a payout end to end.
  2. Review the signing helper against Authentication.
  3. Confirm every write sends an Idempotency-Key that is stored before the request.
  4. Confirm webhook handlers verify the signature on the raw body.
  5. Work through the Live checklist before you switch to a Live key.

What’s next

MCP connectors

Query your account data from an AI app.

Live checklist

Review the full pre-launch checklist.