Planning Award Travel With AI: ChatGPT, Claude, and MCP Connectors

ChatGPT and Claude can help describe destinations, compare broad booking strategies, parse your request, and explain a completed calculation. They cannot reliably plan an award from memory because transfer relationships, award prices, and bonuses change. With the AwardMath connector, either assistant can send your trip to a deterministic solver and return up to three dated funding plans.

Without a connected service, the assistant works from what you type, what it remembers from training, and any web pages it can access. That is useful for questions such as whether a stopover would make a trip easier or which airports serve a region. It is weaker at questions such as which of your bank balances can produce exactly the miles required for a specific program.

AwardMath separates language from calculation. The model passes your words to the same parser used by the AwardMath planner. The parser turns those words into structured balances and trip details. Enumeration and scoring then compare eligible funding paths. The language model explains the returned plans, but it does not choose one or perform the arithmetic itself.

As of August 22, 2026, the underlying graph contains 45 loyalty programs and 260 usable published award-chart rows across 22 programs. Dynamic-priced programs such as Delta SkyMiles, United MileagePlus, and American AAdvantage are excluded from chart solving because their actual prices require a date-specific airline search.

Chatbots get points facts wrong because they predict likely text from imperfect and sometimes old information. A model may recall a former transfer partner, merge two airlines’ rules, or repeat an award price that has since changed. It has no built-in penalty for giving a fluent answer from stale data instead of stopping to verify the relationship.

Training cutoffs are the first problem. Transfer partners, ratios, bonuses, and award charts can change between model updates. A statement that was accurate during training may no longer describe the program you are about to use.

Probabilistic recall is the second problem. A chatbot does not retrieve a partner list the way a database selects rows. It generates the next likely words. Similar program names, shared Avios branding, alliance relationships, and old blog posts can produce answers that sound coherent but combine incompatible facts.

The third problem is that funding an award is a search problem rather than a trivia question. The calculation may need to compare several bank balances, account owners, transfer ratios, bonuses, fees, and chart prices. Reliable planning assigns natural-language interpretation to the model and the combinatorial work to a deterministic solver.

A dated example shows why structure matters. On August 19, 2026, AwardMath’s extractor encountered Turkish Miles&Smiles displaying a 65,000-mile capacity-controlled business saver award between the United States and Türkiye beside a 135,000-mile flexible award. The first extraction mislabeled the flexible price as the saver. The correction added a price-basis field to every award row. A model recalling the page would not automatically receive that structural fix.

MCP is an open standard that lets an AI assistant call outside tools and retrieve structured data during a conversation. Anthropic introduced the Model Context Protocol in November 2024, and OpenAI added MCP connector support to ChatGPT in 2025. The assistant handles the conversation, while the connected service performs the defined work and returns a structured result.

An MCP connector is often described as an MCP server. It publishes named tools with specific inputs and outputs. Instead of asking a model to remember your balances or estimate a transfer path, the model can call the appropriate tool and use the returned data in its answer.

The AwardMath connector shipped in August 2026. It supplies four tools. get_wallet reads saved balances, trips, and watches. plan_award_trip plans against saved balances or balances typed into the conversation and returns up to three plans with a link to view them on AwardMath. save_trip stores a planned trip in My Trips. program_facts returns transfer relationships and published award-chart rows for one program, including observation dates and verification tiers.

The connector does not give the model unrestricted access to AwardMath. It exposes only those typed actions. It never books an award, transfers points, or signs in to a bank or airline.

Create an AwardMath account, save or enter your balances, generate a private connector URL from My Trips, and add that URL to Claude as a custom connector. The account uses a confirmed email and magic-link sign-in with no password. The connector URL contains its own secret key, so store it and share it as carefully as a password.

Open My Trips and find the “Connect an AI assistant” card. Generate a link in the form https://awardmath.com/api/mcp/YOUR-KEY. In claude.ai, open Settings, choose Connectors, and add a custom connector using that URL. The exact placement of platform settings can change, so use the current connector setup instructions if the labels have moved.

Claude Code uses an HTTP MCP connection. The documented command is claude mcp add --transport http awardmath "YOUR-URL". Replace YOUR-URL with the private link generated by AwardMath.

Once connected, Claude can ask AwardMath for your saved wallet, send a trip to the solver, retrieve program facts, and save a returned plan. Claude cannot see card credentials or airline logins because AwardMath does not hold them. You can rotate the connector key or disconnect it from the same My Trips card.

Generate an AwardMath MCP URL from My Trips, then add it in ChatGPT under Settings and Connectors. Developer mode may need to be enabled, and authentication should be set to none because the secret is already carried inside the connector URL. Treat that URL as a password and rotate it if you expose it.

You need an AwardMath account before creating the link. The account uses email confirmation and magic-link sign-in rather than a password. Open the “Connect an AI assistant” card in My Trips, copy the URL, and follow the current ChatGPT connector setup.

After the connection is available, you can ask ChatGPT to read the balances saved in AwardMath, plan a named trip, retrieve facts about one loyalty program, or save a returned trip. The assistant passes your own wording to AwardMath’s parser. A guessed program name generated by ChatGPT does not silently become the mathematical input.

The connection does not let ChatGPT book travel or move points. It also does not provide live airline inventory. It adds dated program facts and funding calculations to the conversation, not control over your loyalty accounts.

A normal ChatGPT or Claude conversation cannot confirm that an award seat exists right now, and the AwardMath connector does not add that ability. Availability is a live, date-specific airline query. Dedicated tools such as Seats.aero, PointsYeah, and Roame search that inventory, but you still need to confirm the result with the ticketing airline before transferring points.

A web-enabled chatbot may find pages discussing typical availability or published award prices. That is not the same as querying the inventory available for your exact flight, date, cabin, and mileage program. Search snippets can also lag behind the underlying airline system.

As of August 19, 2026, Seats.aero said it searches award flight availability across more than 20 airline mileage programs. PointsYeah searches flight and hotel awards. Roame said it searches up to 25 airline loyalty programs as far as 365 days ahead. These products perform the availability-search job that AwardMath intentionally does not.

Use an award-search tool to locate candidate space. Then sign in to the airline program that would issue the ticket and reproduce the search. Confirm the mileage price, taxes, passenger count, and cabin before moving any transferable points. Transfers are generally irreversible.

No. An AI assistant can organize options and explain tradeoffs, but you should decide which balance to spend after checking the seat, cash price, fees, transfer timing, and your future plans. AwardMath’s language model never selects a funding plan. Its deterministic solver returns up to three candidates, and you choose whether any candidate makes sense.

A mathematically possible transfer is not automatically the right redemption. The solver can establish that a balance covers an award and show the dated ratio and chart price used. It cannot know how strongly you value preserving Chase Ultimate Rewards for a future Hyatt stay, keeping Amex Membership Rewards for another traveler, or paying cash to retain flexibility.

The correct role for the assistant is explanation. It can identify why one plan consumes fewer transferable points, why another has lower cash fees, or why an unverified chart row deserves another check. The decision remains yours because value depends on information outside the funding graph.

Program names matter in every verdict. Spending 40,500 Amex Membership Rewards through Iberia Plus is a different decision from spending 89,000 Singapore KrisFlyer miles, even if both could cover the same broad route. The balances left afterward are different assets.

Sharing point totals is lower risk than sharing bank or airline credentials, but the balances are still personal financial information. Use a connector only with an AI account you trust, avoid placing account numbers or login details in prompts, and treat the AwardMath connector URL as a password because anyone holding its key may call the enabled tools.

AwardMath does not store card logins or airline logins, so its connector cannot reveal credentials that it never had. It can read the balances, saved trips, and watches associated with your AwardMath account through get_wallet. It can also save a trip when asked through save_trip.

The private URL looks like https://awardmath.com/api/mcp/YOUR-KEY. Do not post it in a public conversation, screenshot, shared document, or support forum. If you think the URL was exposed, rotate or disconnect it from the “Connect an AI assistant” card in My Trips.

You can also plan without saved balances. The plan_award_trip tool accepts balances typed into the current conversation. That reduces what is stored in your AwardMath wallet, although the information still passes through the AI service whose conversation you are using.

An AI assistant can replace some manual explanation and note-taking, but it does not replace live award search or final airline verification. An AwardMath MCP connection adds dated transfer facts, saved balances, deterministic funding calculations, and trip saving. Seats.aero, PointsYeah, and Roame remain the tools for finding candidate seats because AwardMath does not query availability.

The dedicated-search row covers products with different published features. Roame Wallet tracks cards and point balances on its free tier, while the reviewed Seats.aero and PointsYeah pages focus on availability search. Their public pages do not establish one common trip-saving or wallet capability across the category.

PointsYeah has the stronger documented free search offering and includes hotel awards. Seats.aero is suited to deeper flight searches when you already know the route. The Seats.aero versus PointsYeah comparison covers that choice. If you want optional paid booking help, point.me differs from Seats.aero because point.me sells a separate Concierge service.

AwardMath answers the separate question of which of your balances can fund the award. It does not replace any tool whose main job is finding a bookable seat.

Transfer bonuses and devaluations make undated chatbot answers unsafe because they change the number of bank points required for the same award. A temporary bonus can end before you transfer, while a devaluation can raise the award price without changing the route. Any recommendation should identify the observed ratio date, chart date, and confidence tier used in the calculation.

A model answering from memory may preserve the exciting part of an old promotion while losing its expiration date. It may also combine a former chart price with a newer transfer ratio. The arithmetic can be internally correct and still produce an unusable plan because its inputs never existed at the same time.

AwardMath timestamps each ratio and award row returned by the solver. As of August 22, 2026, the graph’s latest transfer observations included 5 verified pairs, 94 corroborated pairs, and 27 unverified pairs. Verified means the program’s own published page supports the fact. Corroborated means independent sources agree. Unverified means only one source supports it and you should check it directly.

Devaluations are especially difficult for static references. AwardHacker shut down and cited dynamic pricing and reduced availability as reasons its former chart-lookup model no longer worked. The AwardHacker alternatives comparison separates replacement chart tools from live availability search and points-funding calculations.

AwardMath gives ChatGPT or Claude access to your saved wallet, dated program facts, deterministic award-funding calculations, and trip saving through four defined MCP tools. It does not add booking, point transfers, airline logins, card logins, or live seat availability. The connector narrows the assistant’s job to conversation while AwardMath performs the structured lookup and math.

The distinction matters because the model does not need to reproduce the loyalty graph from memory. It can call program_facts for one program and receive transfer partners and chart rows with their observation dates and verification tiers. It can call plan_award_trip with your saved balances or balances typed into the conversation.

The returned plan includes a live AwardMath link where you can inspect it. If you want to keep the trip, save_trip stores it in My Trips. If you do not connect an assistant, the same planning engine remains available through the web planner.

AwardMath is free to use and has no subscription. After a plan is produced, the site asks whether you want to leave a voluntary contribution. Nothing is required, and no planning result is gated behind payment.

Deterministic math evaluates defined inputs through repeatable enumeration and scoring, while a chatbot generates likely language. Give the solver the same balances, trip, transfer data, and chart rows, and it evaluates the same funding paths. The language model may phrase the explanation differently, but it does not alter the underlying calculation or choose the winning plan.

This matters when several balances can reach the same airline program. A casual answer may stop after noticing one transfer partner. A solver can compare all eligible paths in its graph, account for the transfer ratio attached to each path, check whether the available balance covers the requirement, and rank the feasible outcomes.

AwardMath also preserves the provenance of the inputs. Every returned number carries an observed date and verification tier. A verified row came from a program’s published page. A corroborated row was supported by independent sources. An unverified row came from a single source and should not receive the same confidence.

The system still has limits. It does not know whether the airline currently offers the seat, and it excludes dynamic-priced programs such as Delta SkyMiles, United MileagePlus, and American AAdvantage from fixed chart solving. Those prices must come from a date-specific search.

A worked August 22, 2026 calculation used 250,000 Amex Membership Rewards points and 80,000 Chase Ultimate Rewards points for a November 2026 business-class trip from JFK to Madrid. The first plan transferred 40,500 Amex points to Iberia Plus at 1:1, then used the 40,500-Avios off-peak business price plus $117.10.

The Iberia Plus transfer ratio was observed August 19, 2026. The award-chart row was observed May 1, 2025 and carried an unverified tier, so the traveler should verify the price directly. Peak dates can run as high as 59,000 Iberia Avios in the chart data used by the solver.

An alternative funded 89,000 Singapore KrisFlyer saver miles and showed $5.60 in fees. The point difference is substantial, but that does not make Iberia Plus automatically correct. The traveler still needs to compare actual availability, schedule, operating carrier, cancellation terms, transfer speed, and the value of preserving each balance.

Neither result claims a seat exists. The calculation answers whether the recorded balances can fund the published award options. The traveler must find and confirm the seat through the relevant airline program before transferring Amex Membership Rewards or Chase Ultimate Rewards.

Delta SkyMiles, United MileagePlus, American AAdvantage, and similar dynamic-priced programs are excluded because their award cost is determined through a flight-and-date search rather than a stable published chart row. Treating a historical or typical price as the current price would make the funding calculation look precise while using an input that does not apply to the chosen seat.

A fixed chart can assign a price by region, distance, cabin, or route. Once the route and rule are known, a solver can compare that published requirement with your transferable balances. Dynamic pricing allows the airline to vary the required miles by date, flight, demand, or other internal factors.

That makes the current mileage price part of availability search. AwardMath does not perform that search, so it excludes those programs from the chart solver by design. As of August 22, 2026, the solver uses 260 published award-chart rows across 22 programs where a structured price can be represented.

You can still use an AI assistant to discuss a Delta SkyMiles, United MileagePlus, or American AAdvantage result that you found. Enter the observed price as context, but return to the airline before transferring or booking because the price and inventory may change.

Use the assistant to describe the trip, AwardMath to calculate funding paths, a dedicated search tool to find candidate seats, and the airline to confirm and book. This division keeps each system on the task it can perform. Do not transfer points until the final airline search shows the required seats, mileage price, cabin, and taxes.

Start by entering your balances and trip in ChatGPT or Claude through the AwardMath connector, or use the AwardMath planner directly. Review the returned plans, especially the observation date and verification tier attached to each ratio and chart row.

Next, search Seats.aero, PointsYeah, Roame, or the airline site for the relevant route and dates. Seats.aero works well when the route is known and you want a dense flight-results table. PointsYeah covers flights and hotels and positions its core search as free. Roame is built for broader browsing and publishes region views, a mobile app, and searches across up to 25 airline loyalty programs as of August 19, 2026.

Reproduce the chosen result on the airline’s own site. Only after that confirmation should you transfer points. AwardMath verifies the funding math, the search tool identifies candidate inventory, and the airline controls the final price and seat.

Points are not the right payment method when the cash fare is low, the award carries high fees, the transfer would empty a valuable balance, the itinerary is inflexible, or the seat cannot be confirmed. A solver can show that a redemption is possible, but possibility alone does not establish that transferring Amex Membership Rewards or Chase Ultimate Rewards is sensible.

Compare the complete cash cost with the points option. Include taxes, carrier surcharges, positioning flights, cancellation rules, and any points you would earn on a paid ticket. A low mileage requirement can conceal a large cash component, particularly when the issuing program passes along carrier surcharges.

Preserve flexibility when the trip is uncertain. Bank points can often reach several airline or hotel programs, while transferred miles are generally trapped in the receiving program. Moving Amex Membership Rewards to Iberia Plus before confirming a seat converts a flexible asset into a program-specific one.

You should also reject a plan built on weak data when the price cannot be reproduced. The 40,500-Avios Iberia Plus chart row in the August 22, 2026 example was unverified. That makes direct confirmation necessary before the result should influence a transfer.

AwardMath dates every transfer observation and award-chart row so you can see how fresh each input is, but not every fact has the same verification strength. As of August 22, 2026, the graph held 45 loyalty programs, with 5 verified transfer pairs, 94 corroborated pairs, 27 unverified pairs, and 260 usable chart rows across 22 programs.

A verified fact comes from the loyalty program’s own published page. A corroborated fact is supported by independent sources that agree. An unverified fact rests on one source and should be checked before you act. The planner shows the tier rather than presenting all rows as equally certain.

Observation dates also matter. The Iberia Plus ratio in the worked scenario was observed August 19, 2026, while the relevant chart row was observed May 1, 2025. The solver can still calculate with that row, but the older date and unverified status tell you to confirm the award price directly.

No timestamp turns static data into live inventory. Even a ratio checked today does not prove that a specific award seat is available. Confirm both the transfer terms and seat with the relevant program before moving points.

Can ChatGPT find award flights for me?

ChatGPT can help frame the search, but it cannot confirm live award inventory from memory or ordinary web results. Use Seats.aero, PointsYeah, Roame, or the airline site to find candidate seats. Confirm the exact seat and price with the ticketing program before transferring points.

Can Claude access my Amex and Chase points?

Claude can read balances you saved in AwardMath after you add the private MCP connector. AwardMath does not hold Amex Membership Rewards or Chase Ultimate Rewards login credentials. Claude receives the balances exposed by the connector, not access to the underlying bank accounts.

Does the AwardMath MCP connector book flights?

No. The AwardMath connector can read a wallet, retrieve dated program facts, calculate up to three funding plans, and save a trip. It cannot transfer points, book an award, access airline logins, or verify live seats. You complete the final search and booking with the airline.

Is MCP required to use AwardMath?

No. MCP lets ChatGPT or Claude call AwardMath from a conversation, but the same deterministic solver is available through the AwardMath planner. You can type balances naturally or upload a screenshot, correct the returned wallet card, and plan without connecting an outside assistant.

Can AI compare all my transfer options?

A chatbot alone may overlook partners or use stale ratios. With AwardMath connected, the assistant can send your balances and trip to a deterministic solver that compares eligible paths in its dated graph. As of August 22, 2026, that graph covered 45 loyalty programs.

Can ChatGPT tell me how many points a flight costs?

It can quote a published chart price if supplied with reliable, dated data, but it cannot infer the live price of a dynamically priced Delta SkyMiles, United MileagePlus, or American AAdvantage award. Search the exact flight and date, then confirm the result with the airline.

Should I transfer points based on an AI answer?

No. First reproduce the award on the airline’s site and confirm the seats, cabin, mileage price, taxes, and passenger count. Transferable points are generally irreversible once moved. Use AI for explanation and funding math, not as proof that inventory exists or the price is still valid.