Blog
eSIM API for AI agents: connectivity as a tool, not a widget
Travellers still ask for data that works on arrival. For AI travel agents, that should be a tool call—not a deep-link into a third-party storefront mid-conversation.
This page is for builders integrating an eSIM API for AI agents and OTAs, once an agent already holds a booking and a destination.
What the agent must be able to do
- Search —
esim.plans(or equivalent) by country/region, duration, and data need. - Present — supplier-named offers with price, coverage notes, and activation steps the model can explain.
- Fulfill — purchase/activate only after an explicit yes; replay safely with an idempotency key.
- Attach context — tag the call with customer / PNR metadata so you can invoice and support later.
Ancilair’s shelf example uses placeholders where live rates are not public:
esim.plans { country: "FR", days: 7, data_gb: 5 }
Response shape (illustrative): a plan list, which supplier answered, the cost quoted before the order, and cache state.
Multi-ancillary shelf vs single-category aggregators
| Lens | Single-category eSIM aggregator | Multi-ancillary agent shelf |
|---|---|---|
| Auth | Separate API key per category | One token across lounge, insurance, eSIM, … |
| Agent UX | One tool silo; other jobs need other SDKs | Same session: layover lounge + destination eSIM |
| Pricing model | The aggregator’s own price | The supplier’s rate, quoted on the call |
| Ops | Multiple audits / failure modes | Shared budgets, tags, idempotency, playbooks |
Single-category specialists can still be the best supplier behind a routed tool. The product decision for agent builders is whether the integration surface is one catalog or five.
Where eSIM sits in the trip workflow
- After air is booked (or while the offer is held): destination connectivity.
- Beside lounge search on long layovers.
- Beside insurance when the agent packages “ready to travel.”
- Via MCP if your host already speaks tools.
For the broader “plan vs fulfill” framing, see Build an AI travel agent that fulfills ancillaries.
Implementation checklist
- Strict schemas: reject unknown fields before billing.
- Cache policy for identical plan searches (team-level cache can cut repeat cost).
- Document delivery: QR / activation payload via a short-lived host URL if the tool needs files.
- No invented coverage claims in prompts—relay supplier text.
Sources
- Ancilair —
esim.planson the ancillary shelf - Ancilair MCP install patterns — MCP · REST · CLI
Not telecom or consumer advice. Plan availability and fair-use rules are supplier-specific.
Questions
- What does an eSIM API need to expose to an AI agent?
- At minimum: search plans by destination (and often days / data GB), return supplier price and activation constraints, then issue or activate after an explicit traveller confirmation—with idempotent retries.
- How is this different from a consumer eSIM marketplace?
- Marketplaces optimise for human browse-and-buy. Agent APIs optimise for tool schemas, quoted cost before call, credential isolation, and sitting next to other trip tools (lounge, insurance) under one auth model.
- Can we bring our own eSIM wholesale contract?
- Platforms that support bring-your-own credentials let calls run on your supplier key (often unmetered on the marketplace balance). Confirm with your provider’s contract terms.