Agents initiate refunds; refunds enter a pending state that needs explicit authorisation before money moves.
Payments built for AI travel agents.
felloh gives AI agents the booking-aware payment primitives, MCP server and prompt patterns they need to take real bookings, payments and refunds. With safety built in, not bolted on.
Generic payment APIs were not built for agents.
Travel agents do not move money once. They take a deposit, schedule a balance, tokenise a card, sometimes refund, sometimes rebook, and they always sit against the same booking. AI agents inherit all of that complexity the moment they are pointed at a payment API. felloh was built for travel bookings from day one. That same booking-aware ledger is what we expose to AI agents.
Four building blocks for agentic travel payments.
Each block stands on its own, and they compose. Wire an agent against the API directly, drop the MCP server into Cursor or Claude Code, or follow the prompt patterns inside your own framework.
Booking-aware billing primitives for agents.
Agent-scoped endpoints for customers, bookings, payment links, ecommerce sessions, scheduled instalments and refunds. Built around bookings, not generic transactions.
- Agent endpoints for customers, bookings and payment links
- Scheduled payments for deposits and balances
- Two-stage refund authorisation to keep humans in control
Building blocks for the full booking lifecycle.
Felloh handles payment processing, compliance and fund management. Your agent handles conversation and business logic. Workflow patterns cover deposit, balance and refund handling out of the box.
- Deposit, balance and refund workflow patterns
- Tokenised cards for repeat balance collection
- Sandbox environment with test scenarios and credentials
Drop felloh into the agent client your team already runs.
A hosted MCP server exposing 60+ tools across bookings, transactions, refunds, chargebacks, scheduled payments, disbursements and more. No API client code required.
- Works with Claude Desktop, Cursor, VS Code, Windsurf, ChatGPT
- 60+ tools across the full booking and payment lifecycle
- API-key pair auth · the server handles token caching
Battle-tested prompts so agents act safely from day one.
Templates, guardrails and tool-selection patterns drawn from real travel-payment integrations. Prevents the off-by-100 currency errors, invented bookings and unsafe refunds that catch out generic LLM agents.
- System prompts for deposit, balance and refund flows
- Currency unit (minor units) and confirmation guardrails
- Patterns for escalating high-value refunds to a human
Native to the models, frameworks and clients your team ships in.
Native MCP support for Claude and OpenAI, adapters for LangChain and CrewAI, plus drop-in editor support across Cursor, Claude Code, VS Code, Windsurf and ChatGPT.
- Claude Model
- OpenAI Model
- LangChain Framework
- CrewAI Framework
- Cursor Client
- Claude Code Client
- VS Code Client
- Windsurf Client
- ChatGPT Client
One booking, the whole agent flow.
Create the customer, open the booking, send a payment link for the deposit, schedule the balance, and refund through a two-stage authorisation. Every call hits the same booking-aware ledger.
See the full agent endpoint referenceGuardrails travel finance teams can actually sign off on.
felloh assumes agents will sometimes get it wrong. So the platform enforces the controls travel finance teams need before letting an LLM touch real money.
API key pairs limit what an agent can see and do. Rotate or revoke without redeploying the agent.
Prompts and SDKs enforce minor-unit amounts, so an LLM never silently turns £84.20 into £8,420.
Templates require a confirm step before refunds, cancellations or high-value changes.
A fully isolated sandbox with test data, test cards and scripted scenarios for deposit, balance and refund flows.
Put AI to work on your travel payments.
Talk to the felloh team about wiring AI agents into your booking, payment and refund workflows — without giving up the controls your finance team relies on.