Skip to content

Can our support team use Recharge in ChatGPT?

Recharge in ChatGPT lets support look up a subscriber's charges, orders and status. Use a read-only token and keep refunds and cancels restricted.

Uday Gajavalli 6 min read
A support agent asking ChatGPT why a subscriber was charged, with a Recharge charge record and its status beside the chat

Why a subscription support team wants Recharge in ChatGPT

Priya Shah leads support for a coffee brand that sells subscriptions through Shopify and Recharge. Her queue repeats itself. A subscriber says they were charged twice. Another skipped a box and wants to know why the card was hit anyway. Someone else asks when the next shipment goes out. Each answer sits in Recharge, across a charge, a subscription and an order. Recharge in ChatGPT puts those lookups next to the ticket the agent already has open.

MCP (Model Context Protocol) is the standard way an AI assistant calls tools in other apps. Recharge does publish one: the Recharge Analytics MCP, which connects a store's subscription analytics to Claude, ChatGPT or any MCP client (getrecharge.com, checked November 2026). Its support article says it answers aggregate metrics only and cannot list individual subscribers (support.getrecharge.com, checked November 2026). A ticket is about one subscriber, so it needs a connector that calls the Recharge API.

Elaichi's recharge connector is that connector. It signs in with one Recharge API token and carries reads, writes and deletes across charges, subscriptions, customers, orders and plans. A support desk needs only the reads.

Which Recharge token scopes a lookup-only desk needs

Read access on four scopes covers every lookup in this playbook. Recharge authenticates every request with an API token, and scopes set on the token's edit page control what it may reach (developer.rechargepayments.com, checked November 2026).

Only storeowners see the API tokens page by default. To create a token, a storeowner opens the merchant portal, then Tools & apps, then API tokens, then Create an API Token. They set a name and a contact email. Then they choose No access, Read access or Read and Write access for the scopes (docs.getrecharge.com, checked November 2026).

Charges have no scope of their own. Listing charges or orders needs read_orders, and subscriptions, customers and plans each have their own read scope (developer.rechargepayments.com, checked November 2026). So set Read access on orders, subscriptions, customers and plans, and leave every write at No access.

Recharge's write scopes are coarse, which is why the token alone is not the answer for every team. Skipping a charge needs write_orders. Refunding one lists write_orders and write_payments, and Recharge says either is enough (developer.rechargepayments.com, checked November 2026). A token that can skip can also refund. Recharge's own security guide advises starting with read-only scopes and one token per application (docs.getrecharge.com, checked November 2026).

Which Recharge tools answer "why was I charged?"

Five list tools and one schedule read cover the questions in Priya's queue, and each narrows to a single customer. list_all_recharge_customers finds the subscriber by email. list_all_recharge_subscriptions filters by customer and status and returns the price and next charge date. list_all_recharge_charges filters by customer, status and scheduled or processed dates.

That charge filter carries the real answer to "why was I charged twice". Its status values include success, queued, error, refunded, partially refunded and skipped. A charge in error carries the error and its retry date. list_all_recharge_orders shows each order's items, status and scheduled date. list_all_recharge_plans shows a plan's discount and subscription preferences. recharge_customers_get_delivery_schedule answers "when is my next box" for one customer.

One limit trips agents up. Recharge's API stops returning a processed charge once its processed date is more than 90 days old. The data stays in the merchant portal and the Exports tool (developer.rechargepayments.com, checked November 2026). An empty answer about last spring's charge means "look in the portal", not "no charge".

Which Recharge writes the support role should never reach

Every Recharge tool that moves money, ends a subscription or removes a record stays restricted for support. The catalog marks the connector's delete tools destructive, and recharge_subscriptions_cancel is one of them.

Kind Example tools For the support role
Subscriber reads list_all_recharge_customers, list_all_recharge_subscriptions, list_all_recharge_charges, list_all_recharge_orders, list_all_recharge_plans Allow
Schedule read recharge_customers_get_delivery_schedule Allow
Money recharge_charges_refund, recharge_charges_process, recharge_charges_capture_payment, recharge_charges_apply_discount Restrict
Schedule changes recharge_charges_skip, recharge_addresses_skip, recharge_subscriptions_set_next_charge_date Restrict
Cancels and deletes recharge_subscriptions_cancel, update_a_recharge_subscription_by_id, delete_a_recharge_subscription_by_id, delete_a_recharge_customer_by_id Restrict
Customer email create_a_recharge_notification Restrict

Two rows deserve a closer look. update_a_recharge_subscription_by_id accepts a status of active, cancelled or expired, so a plain update can cancel. delete_a_recharge_customer_by_id cancels and deletes the customer's addresses, subscriptions and one-time items with it. Blocking only the cancel tool leaves both paths open.

create_a_recharge_notification sends an email from a Recharge template to the subscriber. That is a message under the brand's name, so it belongs with the agent in the help desk, not with the model.

How to write the rule on the support role

One allow rule on the support role gives Priya's team the reads and nothing else from Recharge. A restriction is a rule naming which connectors and which individual tools a target may reach. In Elaichi, restriction targets are role or user only; the organization default is the absence of a rule, which means allow everything. So the support team needs its own role, and each agent holds exactly one.

Name the six read tools in the rule. The same rule, or a second allow rule on the role, must also name every other app support uses, as whole connectors. In Elaichi, once a role carries any allow rule, every connector and tool none of its allow rules names is blocked for that role. A help desk or Shopify connector left off the list is lost to the role. Do not name the Recharge connector whole and some of its tools in one rule. Elaichi refuses that overlap on a new rule.

Then add block rules on the refund, cancel, skip, update and delete tools. Within a layer, a block always beats an allow. If someone later widens the allow rule to all of Recharge, those tools stay restricted. Block rules match a tool's name or its operation, and why blocks match names and allows do not explains the difference. A restriction change takes effect within about two minutes.

Connecting Recharge once for the whole desk

A support admin connects the recharge connector once with the read-only token, then shares it with the team at use. A shared connection runs on its owner's credential, so every agent's call reaches Recharge with the same token. Elaichi records which agent made each call.

Pick an owner who will stay. When an owner leaves, Elaichi's offboarding lists every connection they own, and an admin transfers each one to another member. A transfer leaves every share as it was. Offboarding when the agent holds access covers the rest of that list.

Every organization uses one address: https://api.elaichi.ai/mcp. In ChatGPT, Elaichi is added as one custom app. On ChatGPT Business, only admins and owners create and publish custom apps for the workspace, and members then connect the published app (help.openai.com, checked October 2026). Each agent signs in with their own OAuth grant, the sign-in that ties their Elaichi identity to ChatGPT. Connecting Elaichi to ChatGPT has the steps. In Elaichi, connected tools are never listed one by one, however few there are. So a client control that gates tools by name can act only on execute_tool and its siblings as wholes, never on one Recharge tool. The support role's rules are the control here.

Checking the rule and the audit trail before rollout

Sign in from a test account on the support role and ask ChatGPT for one subscriber's last three charges. The read should run. Then search for "refund" and "cancel". A restricted tool is left out of the tools Elaichi offers a client and cannot be called. Search still names it, flagged restricted, with no schema.

Then read the audit trail. Elaichi writes one entry for each connected-tool call that reaches execution (a call refused earlier writes none). Each entry records the tool, the connection reached, the outcome and the client, plus the one path argument that names the object, as the target id, and nothing else about the arguments. For a list call, the row records only the resource type and how many items came back, nothing about the subscriber's charges. What an AI audit log must capture lists what to check in those rows.

Mind the rate limits on both sides. Recharge gives each token a bucket of 40 requests that drains at 2 a second, and a full bucket answers 429. Recharge suggests waiting at least 2 seconds before a retry (docs.getrecharge.com, checked November 2026). Elaichi allows 120 MCP requests every 60 seconds per OAuth grant. A shared token means the whole desk shares one Recharge bucket.

When an agent later needs a restricted tool, they file an access request. An admin approves it in the console, under Governance, then Access requests. The grant lifts that one tool for that one person, and the role is not edited.

When the Recharge Analytics MCP or one seat is enough

If the questions are about the business rather than one subscriber, Recharge's own MCP may be all a team needs. Its support article lists data categories such as subscribers, renewals and renewal rates, payment errors and recovery, and cancellation prevention. It is read-only and cannot change subscriptions or orders. Members need the Analytics permission in Recharge, and a Claude owner adds it from Claude's connector directory (support.getrecharge.com, checked November 2026). That is one vendor, no token to manage and nothing to restrict.

Elaichi earns its place when the question is about one subscriber and crosses apps. The charge lives in Recharge, the complaint sits in the help desk and the order detail sits in Shopify. One endpoint serves ChatGPT, Claude, Cursor and Microsoft Copilot alike, with one set of rules and one audit trail. The help desk side of the same desk is in Intercom for support agents in ChatGPT, and payments held to reads are in Stripe access for a finance team. If one person only wants to try a few lookups, you may not need a control plane yet. The rest of the stack is in the 900+ connectors.

FAQ

Frequently asked questions

Does Recharge have its own MCP server?

Yes. Recharge documents the Recharge Analytics MCP, in Early Adopter status, which connects a store's analytics to Claude, ChatGPT or another MCP client (checked November 2026). Recharge says it answers aggregate metrics only, not one subscriber's records, and is read-only. Looking up one subscriber's charges takes a connector that calls the Recharge API, such as Elaichi's recharge connector.

Can a Recharge API token be read-only?

Yes. When a storeowner creates a Recharge API token, they choose No access, Read access or Read and Write access for its scopes. Charges sit under the orders scope, so a token with Read access on orders, subscriptions, customers and plans can look up a subscriber but cannot refund, skip or cancel anything (checked November 2026).

Why can ChatGPT not find an old Recharge charge?

Recharge's API stops returning processed charges, those with status success, refunded or partially refunded, once their processed date is more than 90 days old. Recharge says that data stays available in the merchant portal and through its Exports tool (checked November 2026). An agent checking an older charge needs the portal.

Is blocking the cancel tool enough to stop ChatGPT cancelling a Recharge subscription?

No. In Elaichi's recharge connector, update_a_recharge_subscription_by_id accepts a status of active, cancelled or expired, so an update can cancel too. Deleting a customer cancels and deletes all of that customer's subscriptions. An allow rule that names only reads keeps every one of these out of reach.

How quickly does a new Recharge restriction take effect in Elaichi?

A restriction change in Elaichi takes effect within about two minutes, so test a new rule after that window. Revoking a share is faster: in Elaichi, revoking a share or disconnecting an account takes effect on the caller's next call.

Put AI to work in your own apps

14 days of Black, no credit card. Start with one app and one team.

When the trial ends, nothing is deleted. You move to the free Personal plan, and connections, roles and the audit log stay where they are.

Add one link to your AI app

The same link for everyone on your team.

https://api.elaichi.ai/mcp
  • ChatGPT
  • Claude
  • Cursor
  • Microsoft Copilot