AI plays a different role in every corner of your business . It writes integration code for your developers. Your team works out of Chat GPT or Claude. AI agents also interact with your customers. So why does it fail at payments?
AI coding assistants integrating payment for your business don’t read the right documentation. After you go live, every payment query raised ends up with your team finding answers from your merchant dashboard. And the agent for your product still can’t collect payments.
We built Cashfree for Builders to close these gaps. Three products work as your AI stack for the agentic payments era. Let’s take a look at how each one works.
Agent Skills: Integrate payments in under 7 minutes
It’s 11pm and you’re 3 hours into a Cashfree integration that should have been completed by that afternoon.The order creation worked. The checkout rendered. But suddenly the webhook failed and your AI coding assistant in the editor suggests a header name that doesn’t even exist. Your assistant isn’t faulty, it just hasn’t read Cashfree’s docs.
Agent Skills is a CLI that loads Cashfree’s complete product knowledge into your AI coding assistant in just one command. This helps your assistant write the correct code on the first try. Node 18 or later is the only requirement. After you run the installation code, you get an interactive prompt that allows you to choose your preferred AI coding assistant.
A skill file tree and manifest is installed into your project. The skill files cover each and every one of Casfree’s products. Payment Gateway, Payouts, Subscriptions, Secure ID, Cross Border, Settlements & Reconciliation and Autocollect. For instance, Payment Gateways breaks down further into sub skills for backend, mobile & web SDKs, webhooks, refunds, disputes, payment links, token vault, easy split and offers. And alongside the product skill, you also get a getting started skill, validation and testing skill, a debugging guide and migration skills to help businesses moving from other PG’s to Cashfree.
Every skill comes in 2 halves: the happy path and the deep end. SKILL.md is the former, it covers when to use the product, the core workflow, security rules, testing, and a quick diagnostic. references/REFERENCE.md is the deep end that covers full field schemas, per language SDK code, webhooks, payloads and edge case troubleshooting. Your AI coding assistant reads the happy path first and only refers to the deep end when the question requires field level detail. This makes integrating payments both faster and on point. You don’t have to compromise anywhere.
The manifest tells the assistant where to look. It’s written exactly where your AI coding assistant reads project instructions. For Claude Code, that’s Claude.md. For Cursor, its .cursor/rules/cashfree.mdc.
Antigravity, VS Code Copilot, Gemini CLI and the rest also get each of their own.
The manifest maps your requirement to the right skill file. For instance, it points a new integration to the getting started skill. It also runs the validation skill to ensure that the code written by your AI coding assistant is correct. Integration happens in minutes and you get to go live faster.
MCP Server: Run payment operations from your AI client
A business has a hundred payment questions to answer in a day. A customer’s payment failed. The finance team needs to check today’s settlement amount. Refunds have to be processed. Payment links need to be shared with your customer in the next 5 minutes. All of these tasks are handled by your team everyday.
None of them are hard but they pull you away from what you were working on. MCP is a remote server built to solve this problem. It connects your live Cashfree account to the AI clients like Claude that your team already uses. Simply ask your AI client the question and it answers with real figures from your dashboard.
Cashfree’s MCP connects to Claude, Chat GPT, Cursor and Claude Code. Connect your AI client with URL and login. Access is only granted to the client after you approve it from your merchant dashboard.
You can ask your client to pull up settlement details, check on payment status, look into payment failures and more and get answers from your account.
Cashfree’s MCP doesn’t stop with letting your AI client find answers for your team. It also enables your AI clients to execute tasks for you. In order to do so, it has two separate sets of tools. The merchant dashboard tools cover transactions, settlements and reports. They enable your AI client to pull a single transaction’s complete life cycle, look up orders and check which payment modes are currently active or check up on your cashflow and put together reconciliation reports and summaries.
Payment Gateway tools create and fetch orders with address, cart and customer details. They can instantly create or cancel payment links, process refunds and act on your command. Concerns are flagged and resolved before they cost your business a sale.
Your team can now continue working on your AI clients and simultaneously follow up on payment queries that arise. Log in once and just ask.
Agent Toolkit: Build AI agents to collect payments.
While Agent Skills and MCP serve the teams behind building and running the business, Agent Toolkit serves the product your shipping. It is an npm package that exposes Cashfree’s APIs as function calling tools.
If you are building an AI agent, a shopping assistant or support bot or anything else that talks to your customers, our Agent Toolkit enables them to handle payments end to end. It is a curated package that plugs into three highly used frameworks today: Open AI, LangChain and Vercel AI’s SDK. Install it in the framework you prefer and the agent can now collect payments.
There are over thirty things the Agent Toolkit enables your agent to do. It can create and manage orders, collect payments through different payment methods and issue refunds and check on them. Save a customer’s payment details for later and verify your customer’s identity using PAN, GSTIN, bank account, IFSC, name checks, penny drop, and a full KYC including video.
You choose which tools each agent gets. An agent answering “where is my order” will only need the get order tools. The same logic applies to environments. Test mode and production are one setting apart. Build it in the sandbox and switch to production when you’re ready for real customers.
How your business runs from here
The first thing you’ll notice is how fast you go live. An integration that used to eat a week is done in an afternoon, and the bugs that usually show up on launch night get caught before they’re written. Then the dashboard stops being a place your team visits everyday. Every payment question can be answered from your AI client and any payment link, refund or report you need is created or processed in minutes. And the AI agent you’ve been building goes from demo to product. It can finally collect payments.
Installing Cashfree for Builders
The tools are open and versioned. Agent Skills lives at github.com/cashfree/agent-skills, the Agent Toolkit at github.com/cashfree/agent-tools-js. You can read what gets installed, pin versions, file issues, and contribute.
Agent Skills: run npx @cashfreepayments/agent-skills add skills in your project root, pick your assistants, and restart the editor.
MCP Server: add https://mcp.cashfree.com/mcp as a connector in your AI client and authenticate with your merchant account.
Agent Toolkit: npm install @cashfreepayments/agent-toolkit, generate API keys from the Merchant Dashboard, and start in the sandbox.
