Building on Midtrans with AI

📘

Agent skills is still in beta

Have any feedback? We would love to hear from you - let us know here.

AI coding agents can already read the Midtrans docs — but docs tell an agent what the API accepts, not how a payment integration should behave. The official Midtrans agent skill adds that layer: it turns the coding agent you already use into a Midtrans integration engineer that checks merchant readiness before writing code, picks the right product for each payment method, and proves the result in sandbox before anything goes live.

The skill is free, open source, and needs no credentials — it calls no Midtrans APIs by itself. Your agent reads it and applies it to your codebase.

Why integrate with the skill

• Reach your first sandbox payment in the first working session. The skill front-loads the decisions that normally consume integration days: which Midtrans product fits each payment method, the exact signature recipes, which of the seven things called a "callback" is actually your fulfillment signal, and how to model payment state so a refresh or a failed provider call never strands an order.
• Skip the classic payment bugs. The skill encodes the mistakes every integration rediscovers the hard way — reformatting gross_amount before hashing the webhook signature, treating Snap's onSuccess as proof of payment, retrying refunds without an idempotency key, mixing Snap and Core API hosts. Your agent avoids them on the first pass instead of debugging them in production.
• Get proof, not "looks done". The skill ships deterministic signature verifiers, signed webhook fixtures, and a webhook replay tool, so your agent demonstrates that signature verification works and your handler is idempotent — locally first, then against sandbox — and tells you exactly what remains unproven.
• Use the agents your team already uses. One skill folder works across Claude Code, OpenAI Codex, GitHub Copilot, Cursor, OpenCode, and any agent that reads Markdown skills.

What it unlocks

Hybrids are first-class: the skill helps your agent assign an explicit product owner per payment method — for example Snap for cards and OTC, BI-SNAP for QRIS and VA — instead of forcing everything through one product.

Install the skill

With the skills CLI (works with Claude Code, Codex, Cursor, and other supported agents):

npx skills add https://github.com/veritrans/midtrans-agent-skills --yes

Or copy the folder manually from the veritrans/midtrans-agent-skills repository:

Manually copied skills do not auto-update — refresh the folder before major payment work.

How development works

Use integrate-midtrans-payments to add Midtrans Snap checkout to this app.

From that one prompt, a correctly loaded skill walks your agent through:

1. Merchant readiness preflight. The agent confirms account/MID state, sandbox credentials, activated payment methods, dashboard callback URLs, and what proof you expect — inferring what it can from your repository and asking only for missing blockers.
2. Product routing. Each requested payment method is mapped to its Midtrans product before any code, so Snap, Core API, and BI-SNAP request shapes, auth headers, and status semantics never get mixed.
3. Code that fits your codebase. The agent inspects your existing order, payment, webhook, and environment boundaries first, then implements an explicit payment state model with idempotent, replay-safe webhook handling.
4. Verification. Deterministic local checks (signature fixtures, webhook replay) come first, then a sandbox smoke test — and the agent reports exactly which proof level you have reached.

More prompts to try:

Use integrate-midtrans-payments to debug why Midtrans webhooks do not update orders.
Use integrate-midtrans-payments to review our BI-SNAP GoPay tokenization implementation before launch.
Use integrate-midtrans-payments to plan a sandbox-to-production cutover checklist for our stack.

Check that the skill is loaded

Two checks, ten seconds:

1. Files: the skill folder exists at your agent's skill path (for Claude Code: .claude/skills/integrate-midtrans-payments/SKILL.md).
2. Behavior: ask your agent Use integrate-midtrans-payments to plan adding Midtrans payments to this project. A loaded skill answers with a merchant readiness summary — account/MID, sandbox access, payment methods, callback URLs, expected proof — and asks for missing facts before proposing code. If the agent instead jumps straight into generic payment code, the skill is not loaded: re-run the install, and for Cursor-style tools confirm the rule pointing at SKILL.md is in the agent's always-loaded instructions.
   

These are how it would look like in different AI Agents

Plain text docs

Every page on this documentation site has a Markdown variant: append .md to the URL, for example https://docs.midtrans.com/docs/snap-snap-integration-guide.md. Markdown pages cost fewer tokens than HTML and keep the document structure intact.

The complete page index lives at https://docs.midtrans.com/llms.txt, following the llmstxt.org convention. Point your agent there first; the agent skill treats it as the source of truth and re-reads it on every engagement.

Found a misroute or a gap in the skill? File an issue at github.com/veritrans/midtrans-agent-skills.