Stripe's MCP server lets Claude Code look up Stripe's documentation, read your account and create objects such as products, prices and payment links while you build. Stripe's documentation says that after November 30, 2026, its MCP server will stop accepting full-access secret keys and restricted keys without the Agent tag. If you set up Stripe MCP months ago by pasting a key into a config file, that connection will stop working in December. This guide is for a founder building alone with Claude Code: how to connect Stripe the way Stripe now recommends, why to start in a sandbox, how to find old keys in your agent config files, how to migrate, and how approvals work when you are the only person on the account. It is useful without a Soloop account and is not a security audit.
The short answer
Install the Stripe CLI and run stripe agent setup. It detects Claude Code and installs Stripe's plugin, which configures the hosted MCP server at mcp.stripe.com and Stripe's skills. Start Claude Code, run /mcp and sign in to Stripe with OAuth. On Stripe's consent page, grant a sandbox first and add live mode only when you need it.
If any of your tools sends an sk_ secret key or an rk_ restricted key without the Agent tag to mcp.stripe.com, replace it before November 30, 2026: reconnect with OAuth, or create an agent API key with only the permissions the job needs and load it from an environment variable. Keep refunds and subscription cancellations behind an approval, and never tell Claude Code to always allow Stripe's write tool.
What changes on November 30, 2026
Stripe's MCP server supports two ways to sign in: OAuth, for a client acting on behalf of your Stripe user account, and agent API keys, for an AI application that needs independent, persistent access. An agent API key is a restricted key tagged as belonging to an autonomous agent when you create it. It authenticates like any other restricted key, but it is automatically subject to Stripe's approval rules.
Starting in December, requests to the MCP server that use a full-access secret key or a restricted key without the Agent tag receive a 401 response with an OAuth discovery challenge. Stripe's notice is about keys sent to its MCP server. It does not say that the keys your application uses to charge customers change; a key your app uses on its own server is a separate question, covered by Stripe's general advice to prefer restricted keys.
"Migrate by November 30, 2026. Starting in December, Stripe MCP will stop accepting full-access secret keys and restricted API keys without the Agent tag. Replace these keys with newly created agent API keys or reconnect with OAuth."
Stripe docs: Model Context ProtocolUse OAuth for interactive MCP clients; it lets you authorize and revoke a client without sharing an API key. For autonomous clients that cannot use OAuth, create an agent API key with only the permissions the client needs, test it in a sandbox, then expire the old key.
Stripe docs: Model Context ProtocolTo tag a restricted key as an agent key, select "Authorizing agent access to your account" when Stripe asks for the key's intended use. Agent-tagged keys are subject to approval rules; Stripe adds default rules when you create the first one.
Stripe docs: API keysThe @stripe/mcp and @stripe/agent-toolkit packages are deprecated and "will receive no further updates, including security fixes"; Stripe points to the hosted server instead.
GitHub: stripe/aiWhich access fits which job
Most founders using Claude Code at their own keyboard need OAuth and nothing else. An agent key is for work that runs without you, such as a scheduled job or a headless agent in CI.
| Dimension | Fits when | After November 30, 2026 |
|---|---|---|
| Plugin or /mcp with OAuth | You work in Claude Code, Codex or Cursor and can sign in through a browser. You choose sandboxes or live accounts and permissions on Stripe's consent page. | Recommended by Stripe. Revoke a session under OAuth sessions in your Stripe user settings. |
| Agent API key in an environment variable | The agent runs without you, or the client cannot do OAuth. The key carries only the permissions you set. | Accepted. Sensitive actions follow your approval rules. |
| Secret key (sk_) or untagged restricted key (rk_) sent to mcp.stripe.com | Older setups that pasted a key into a config file. | Rejected with a 401. Replace before the cutoff. |
| Local @stripe/mcp package with --api-key | Older tutorials started it with npx and a secret key. | Deprecated with no security fixes. Move to the hosted server. |
Set up Stripe in Claude Code, sandbox first
A sandbox is an isolated Stripe test environment: no card network processes anything and its objects never mix with live mode. Building in a sandbox means a confused prompt creates a test price, not a real one.
- Install the Stripe CLI (version 1.43.3 or later) with npm install -g @stripe/cli or Stripe's install script.
- In your project directory, run stripe agent setup --client claude-code. Run stripe agent setup --status at any time to see what is installed.
- Start Claude Code and run /mcp. Choose the Stripe server and sign in. On Stripe's consent page, grant a sandbox and the permissions you want; leave live mode out for now.
- If you administer the account, open the Dashboard settings for MCP and CLI access and decide separately whether live mode and sandboxes allow MCP at all.
- Keep Claude Code's tool approval on, so each Stripe call that changes something waits for you.
- Ask for a read-only task first, such as "List the products and prices in my Stripe sandbox", and check that the answer matches the Dashboard.
- Before you accept real payments, follow Stripe's go-live checklist. Grant live mode to the MCP connection only for the tasks that need it.
Find old Stripe keys in your agent config
Run these from a terminal, not through an agent, so key values stay out of an AI transcript. Each command prints file names or commit IDs only. They were tested with grep and git on macOS; on Linux, the Claude Desktop config lives elsewhere.
- User-level agent configs holding a Stripe key: grep -lE '(sk|rk)_(live|test)_' ~/.claude.json ~/.cursor/mcp.json ~/.codex/config.toml ~/Library/"Application Support"/Claude/claude_desktop_config.json 2>/dev/null
- Project configs in each repository, from its root: grep -rlE '(sk|rk)_(live|test)_' .mcp.json .claude .cursor .vscode 2>/dev/null
- Which configs use Stripe MCP at all: grep -rlE 'mcp\.stripe\.com|@stripe/mcp' ~/.claude.json .mcp.json .claude .cursor .vscode 2>/dev/null
- Did a key ever reach git history: git log --all --oneline -G '(sk|rk)_(live|test)_' lists every commit that added or removed one, including in files you have since deleted.
- Open each file the commands name and look for a Stripe key in an MCP server entry, an Authorization header, an --api-key argument or a saved command permission.
Migrate before the cutoff
- For each connection the search found, decide: can you sign in with OAuth? If yes, remove the key from the config and run /mcp to reconnect.
- If the job must run without you, create a new key on the API keys page: Create restricted key, choose "Authorizing agent access to your account", and grant only the resources the job reads or writes. Create it in a sandbox first.
- Store the key in your secrets manager or an environment variable, never in .mcp.json itself, and reference it by name (template below).
- Run one read-only request and confirm it succeeds. A 401 with an OAuth challenge means the key is not tagged as an agent key.
- Expire the old key, or rotate it with a grace period if your app also uses it. Stripe lets both keys work for up to 7 days during a rotation.
- Remove any @stripe/mcp entry started with npx and replace it with the hosted server.
Copy the .mcp.json for an agent key
Use this only when OAuth is not possible. Save it as .mcp.json in the project, set STRIPE_AGENT_KEY in your shell or secrets manager before starting Claude Code, and keep .env files and the key itself out of git. The variable name is an example; Stripe's documentation uses AGENT_API_KEY.
Grant only what the job needs
Illustrative examples, not Stripe's defaults: decide the job first, then the permissions. Restricted and agent keys let you set each resource to none, read or write.
| Dimension | Grant | Leave out |
|---|---|---|
| Answer "why did this payment fail?" | Read access to customers, charges, payment intents and invoices. | All write access. |
| Set up products, prices and payment links while building | Write access to products, prices and payment links, in a sandbox. | Live mode until you launch; refunds and payouts. |
| Weekly revenue summary | Read access to subscriptions, invoices and balance. | Customer updates and all write access. |
| Prepare refunds for your review | Write access to refunds, with the default approval rule kept on. | Payouts, transfers and bank account changes. |
Approvals when you work alone
There are three separate gates, and they do different jobs. Claude Code's own permission prompt asks before a tool runs. Stripe's Human confirmation applies to some actions through OAuth, such as creating a refund: Stripe gives your agent a link, you approve in the Dashboard within 24 hours, then you tell the agent to retry. Stripe's approval rules apply to agent-tagged keys.
Stripe's default rules for agent-tagged keys hold refunds, subscription cancellations, payouts, outbound payments and transfers, and currency conversions for review. The agent receives an approval_required error and Stripe opens a request; you approve or deny it in the Dashboard or the Stripe mobile app, and requests expire after 14 days. Two-party approval normally needs a reviewer other than the requester, so on a one-person Stripe account you can save such rules only for actions taken by agent-tagged keys. That is another reason to give an unattended agent its own agent key rather than your own login.
| Dimension | What it does | Suggested setting |
|---|---|---|
| search_stripe_documentation, stripe_api_search, stripe_api_details | Search Stripe's docs and API reference. | Allow. |
| stripe_api_read, get_stripe_account_info | Read objects in the connected account, including customer data. | Allow in a sandbox; ask in live mode. |
| stripe_api_write | Any create, update or delete call, from a price to a refund or a subscription cancellation. | Always ask. Never choose "don't ask again". |
| Treasury tools such as send_money | Move money from a Treasury account. | Leave unused unless you run Treasury; Stripe requires Human confirmation. |
Copy the migration worksheet
Use one copy for your machine and each repository. Note the connection and its new access so you can revoke it cleanly later.
Where Soloop fits
Soloop does not connect to Stripe, read your Stripe data or manage API keys. Use Stripe's tools and Dashboard for everything above.
Soloop is for what comes after you can take payment: according to the Soloop manual, it can research a market with sources, write and publish X posts and replies through a connected account, send outreach through a connected Gmail account and prepare Google Ads search campaigns after setup and payment, with actions waiting for your approval by default. Do not paste Stripe keys into any AI tool's prompt, including Soloop's.
Sources, limitations and refresh
Stripe's Model Context Protocol, API keys, approvals, agent plugin, agent setup and skills documentation and the stripe/ai README were read on November 11, 2026. Stripe has not published a more precise cutoff time than "starting in December". The commands, permission sets and tool settings are editorial guidance tested on a sample directory, not a security audit or a statement from Stripe or Anthropic.
This guide will be reviewed by February 11, 2027, or sooner if Stripe changes the cutoff or how its MCP server authenticates.
Default rules for agent-tagged keys cover refunds, subscription cancellations, outbound payments and transfers, payouts and currency conversions. Requests expire after 14 days. Single-user accounts can save two-party rules only with the actor condition set to agent-tagged API keys.
Stripe docs: Two-party approvalsstripe agent setup detects Claude Code, Codex and Cursor and installs the relevant plugin, skills and MCP server; --client limits it to one tool and --status shows what is installed.
Stripe CLI: agent setupThe Soloop manual describes research with sources, X posting through a connected account, Gmail outreach and Google Ads search campaigns after setup and payment; approval comes first by default. It does not connect to Stripe.
Soloop product manual