gem-hunter
主要言語: TypeScript
Let your agents spend on your behalf. Your payment credentials are never exposed. You approve every purchase.
Link CLI lets agents get secure payment credentials from a Link wallet for one-time purchases and recurring payments — without storing your real card details.
The CLI can produce one of three credential types:
For now, this is only available to US and Canadian Link accounts.
Documentation:
Tip
If you're looking to integrate the Link CLI into your product or agent for consumers to pay with, be sure to look at Integrating into your agent.
npm i -g @stripe/link-cli
Agent and plugin integrations, including the MCP server, run the installed link-cli binary. They intentionally never download or resolve packages at startup, so install the CLI first.
Install the skills:
npx skills add stripe/link-cli
By default when called from an agent (non-TTY), all commands use toon output — a compact, LLM-friendly text format. All commands accept --format [format] for structured output. Other formats: json, yaml, md, jsonl.
List available commands:
link-cli --llms-full
Get a command's full schema with --schema. Example:
link-cli spend-request create --schema
Link CLI can run as a local MCP server. Add the following to your MCP client config (.mcp.json, etc.)
{
"mcpServers": {
"link": {
"command": "link-cli",
"args": ["--mcp"]
}
}
}Alternatively, use serve to expose link-cli as an MCP endpoint over HTTP.
link-cli serve # binds 127.0.0.1:54321 (loopback only) link-cli serve --port 8080 link-cli serve --host 0.0.0.0 # expose beyond localhost (see warning below)
The server handles POST /mcp and GET skill discovery at /.well-known/skills/index.json and /.well-known/skills/{skill-name}/SKILL.md, with OPTIONS preflight support. Other paths return 404, unsupported methods return 405, and malformed or ambiguous request paths return 400. It binds to 127.0.0.1 by default so only the local host can reach it. Anyone who can reach the port can use this CLI's authenticated Link session, so only override --host on a trusted, isolated network — doing so prints a warning.
Run a guided onboarding and demo flow:
link-cli onboard
The link-cli requires a Link account. You can log in to your existing one or create one online.
link-cli auth login
You receive a verification URL and a short phrase. Visit the URL, log in to your Link account, and enter the phrase to approve the connection.
By default, login requests the userinfo:read and payment_methods.agentic
scopes plus the four financial-insights source actions. To leave out financial
data access, use --no-financial-insights; to leave out payment access, use
--no-agentic-payments.
link-cli user-info retrieve --format json
In addition to identity fields, the response can include the user's address, spend limits, and verification requirements:
{
"email": "jane@example.com",
"name": "Jane Doe",
"address": {
"line1": "510 Townsend St",
"line2": null,
"city": "San Francisco",
"state": "CA",
"postal_code": "94103",
"country": "US"
},
"agent_wallet_spend_limits": {
"per_transaction": { "limit": 50000 },
"daily": { "limit": 500000, "used": 120000, "remaining": 380000 },
"thirty_day": { "limit": null, "used": 600000, "remaining": null }
},
"agent_wallet_verification_requirement": {
"status": "identity_verification",
"action_url": "https://app.link.com/finish_setup?verify=identity_verification&intended_email=jane%40example.com&fromEmail=jane%40example.com"
}
}The address field is omitted when its enrichment is unavailable and is null
when the user has no Person record. Finite spend-limit values are cents because
this response does not include a currency. A null limit or remaining amount
means unlimited. agent_wallet_verification_requirement.status can be
not_required, ssn_verification, credit_file_match_ca,
identity_verification, doc_upload_ca, pending, contact_support, or
complete. Its action_url is null when no user action is available.
Retrieve the rules that grant the current app authority to create spend requests without manual approval:
link-cli approval-policy retrieve --format json
Each rule includes an action, a per-purchase limit, and optionally an ordered list of allowed payment method IDs. The API returns an error when no approval policy has been configured.
link-cli payment-methods list
Returns the cards and bank accounts saved to your Link account. Use the id field as payment_method_id in the next step. If you have no payment methods, add new ones in Link.
The list can also include a Link balance payment method with its available balance when that amount is available.
Retrieve one payment method by ID:
link-cli payment-methods retrieve <payment-method-id>
The response contains the same redacted fields as the matching list item, including capability eligibility when available.
Set or change a payment-method nickname:
link-cli payment-methods update <payment-method-id> --nickname "Work card"
Clear a nickname by passing an explicit empty string:
link-cli payment-methods update <payment-method-id> --nickname ""
Set a payment method as the wallet default, or clear its default status:
link-cli payment-methods update <payment-method-id> --is-default true link-cli payment-methods update <payment-method-id> --is-default false
You can pass --nickname and --is-default together. Supply at least one of
them; omitted fields stay unchanged. Setting a new default replaces the previous
wallet default. Link trims surrounding nickname whitespace and returns the
updated, redacted payment method.
link-cli shipping-address list
Returns the shipping addresses saved to your Link account. The response preserves nullable nickname, address, and address fields exactly as returned by the API.
Create a spend request with merchant details, line items, and amounts. If --payment-method-id is omitted, your default payment method will be used, or the first eligible one if no default is set:
link-cli spend-request create \ --payment-method-id csmrpd_xxx \ --merchant-name "Stripe Press" \ --merchant-url "https://press.stripe.com" \ --context "Purchasing 'Working in Public' from press.stripe.com. The user initiated this purchase through the shopping assistant." \ --amount 3500 \ --line-item "name:Working in Public,unit_amount:3500,quantity:1" \ --total "type:total,display_text:Total,amount:3500" \ --request-approval
The --request-approval flag triggers a push notification to the user for approval. Interactive mode polls until the request leaves the approval waiting states. Agent mode returns a spend-request retrieve command to wait for a status change, including a transition to submitted.
Easily approve requests with the Link app.
The approved spend request includes a card object with number, cvc, exp_month, exp_year, billing_address, and valid_until. Enter these into the merchant's checkout form.
link-cli spend-request retrieve lsrq_001
By default, retrieving a spend request doesn't include card details. Pass --include card to see unmasked card details.
To avoid leaking card credentials into agent transcripts or logs, use --output-file to write the full card to a secure local file while stdout shows only redacted data (brand, last4, expiry):
link-cli spend-request retrieve lsrq_001 --include card --output-file /tmp/link-card.json --format json
The file is created with 0600 permissions. If the file already exists, the command fails unless --force is passed. When --output-file is set, the JSON output replaces the card object with redacted fields and adds a card_output_file path.
For agent polling, pass --interval and optionally --max-attempts:
link-cli spend-request retrieve lsrq_001 --interval 2 --max-attempts 300
Link is already integrated with the following agents:
Link CLI can also read a consumer's financial data -- transactions, balances, and connected account details. Agents can use these features to answer personal finance questions and track trends. Financial Insights are powered by Financial Connections, covering 12,000+ US financial institutions.
Financial Insights also provides pre-computed and user-provided insights to power better shopping experiences through the insights commands.
auth login requests all four financial-insights source actions by default:
read_link_transactions, read_external_transactions, read_balances, and
read_source_details. The user decides whether to approve the connection;
requesting access does not retrieve financial data.
link-cli auth login --client-name "My Agent"Use --no-financial-insights for a payment-only login, or
--no-agentic-payments for a login without the default payment scope. To
request a narrower set of financial-data actions, pass the desired
--source-actions flags or --authorization-detail entries; either replaces
the four-action default. Don't combine either option with an explicit
--financial-insights or --no-financial-insights flag. An explicit --scope
replaces the default scopes. If already authenticated, use
auth upgrade --source-actions <action> to add only the access needed without
dropping existing access.
Discovering available insight types only requires an authenticated session. Retrieving an insight can require additional source actions, which the discovery response identifies.
link-cli insights list-available-types --format json link-cli insights list --insight <insight_id> --format json
The first command lists available insight IDs, descriptions, and any access needed to retrieve them. The second returns the selected insight's status and data. Omit --insight to list all insights, or repeat it for several IDs to request only a few. Both commands support --limit and --starting-after to retrieve paginated results. If the provided access token is insufficient to view an insight, a no_data result will include authorization_remediation for missing permissions. A pending result means the insight is still being computed.
link-cli sources list
Returns connected financial accounts (bank accounts, credit cards, etc.) with metadata, capabilities, and connection status. Use the id field as --source in other commands.
link-cli transactions list
Supports server-side filtering:
link-cli transactions list --start-date 2026-01-01 --end-date 2026-01-31 link-cli transactions list --category groceries link-cli transactions list --origin external_connection link-cli transactions list --source <source_id>
| Flag | Description |
|---|---|
--start-date |
Only transactions on or after this date (YYYY-MM-DD) |
--end-date |
Only transactions on or before this date (YYYY-MM-DD) |
--category |
Filter by transaction category |
--origin |
link (Link-native) or external_connection (from linked bank/card) |
--source |
Filter by source ID (repeatable for multiple accounts) |
--limit |
Max results per page (1–100) |
Amounts are integers in the currency's smallest unit. Negative = money leaving the account, positive = money entering. Transactions may be Link-native (processed directly through Link), or sourced through an external connection (e.g. imported from transactions that would appear on a bank statement).
The financial-insights skill teaches agents which command to run for each question type, how to handle pagination, interpret amounts, and summarize results. See skills/financial-insights/SKILL.md for the full agent guide.
link-cli balances list link-cli balances list --source <source_id>
Returns current balances for connected accounts, including cash.available (bank/savings) or credit.used (credit cards).
link-cli auth login --client-name "Claude Code" # identify the connecting agent link-cli auth login --client-name "Claude Code" --interval 5 --timeout 300 # login + poll in one call link-cli auth upgrade --scope "userinfo:read" # widen access to a superset link-cli auth status # check auth status link-cli auth logout # disconnect
When you provide --client-name, the Link app displays it when you approve the connection — for example, Claude Code on my-macbook instead of link-cli on my-macbook.
With --interval, the login command yields the verification code immediately and then polls inline until authenticated or time out — no separate auth status call needed. This is recommended for agents that cannot relay the code while a separate polling command blocks their I/O channel.
auth upgrade accepts the same explicit access flags as auth login, but does
not add the login-only financial-insights default. Use it to widen an existing
session. auth login stops with an "already logged in" message when a valid
session exists; auth upgrade merges the requested access with the current
scope and authorization_details and starts approval for the superset.
If there is no valid session, auth upgrade warns and continues with only the
requested access. An existing session remains valid during approval and is
replaced only after the new grant is approved, so abandoning the approval leaves
the old session working.
auth status reports the scope and authorization_details the current session was granted (echoed by the token endpoint at login/refresh and stored in the credential file), and includes an update field when a newer version is available:
{
"authenticated": true,
"scope": "userinfo:read payment_methods.agentic",
"authorization_details": [{ "type": "source", "actions": ["read"] }],
"update": {
"current_version": "0.1.2",
"latest_version": "0.2.0",
"update_command": "npm install -g @stripe/link-cli"
}
}scope and authorization_details are only present when the token endpoint returned them.
Set NO_UPDATE_NOTIFIER=1 to suppress update checks (for example, in CI).
All commands accept --auth <path> to store auth credentials in a specific file instead of the default location. auth login writes to this file; all other commands read from it. Useful for running multiple sessions with separate identities.
Services can verify these credentials with Agent Identity, a separate SDK in this repository. The event-registration example shows site access using an attestation and registration using a verified email.
Identity commands require Link Agent Wallet v0.27.0 or later for the flow documented here. Check link-cli --version; install or update with npm install -g @stripe/link-cli@latest. They are available in --help and --llms without a preview flag, but remain excluded from MCP. See wallet setup and the identity skill.
Both identity request commands save their artifacts to disk and return only the file path and metadata. This applies to every output format, including JSON, piped output, and --full-output. Request output never includes credentials, tokens, or claim values.
Privacy-preserving attestations that let a service verify Link issued the token, without disclosing personal details or identifying the agent:
link-cli identity attestations request --count 10
Each request adds tokens to the CLI-managed pool at ~/.link-cli/identity/attestations/pool.json. Pop one token when you need to answer an attestation challenge:
link-cli identity attestations pop --format json
pop removes one token before returning its token, ready-to-use authorization header, issuer, and issuer-key ID. Pass authorization as the Authorization header in your browser automation or HTTP client. An empty pool returns ATTESTATION_POOL_EMPTY; refill it with request --count 10.
For agent-managed tokens, export a batch to a new file outside the CLI storage directory:
link-cli identity attestations request --count 10 --output-file ./aats.json
Exported tokens never enter the CLI pool. The agent owns their consumption and cleanup. Existing exports remain separate and are never automatically imported. Wallet credentials keep their existing storage and behavior.
Pool updates are serialized and saved atomically. A crash after removal can lose a token; pop never returns it to the pool. If a crash leaves pool.json.lock, ensure no attestation commands are running before removing that lock directory.
User info that has been signed, proving it comes from Link:
link-cli identity credentials request
identity credentials request saves a signed credential to ~/.link-cli/identity/credentials/current.json, bound to the CLI-managed holder key at ~/.link-cli/identity/holder-key.jwk. Structured output includes output_file, issuer, expiry, holder-key path/thumbprint, and claim names. Use identity credentials present to sign selected claims, capture its JSON output programmatically, and send the presentation through browser automation or an HTTP client without printing it into the agent transcript.
Presentations disclose selected claims to a verifier using the saved credential and holder key:
link-cli identity credentials present \
--aud https://directory.example \
--nonce '<nonce-from-challenge>' \
--claim email \
--format json{"presentation":"<issuer-jwt>~<email-disclosure>~<key-binding-jwt>"}Use the verifier challenge's exact audience and nonce. Repeat --claim to disclose additional claims, such as --claim email --claim email_verified; at least one claim is required. The command reads ~/.link-cli/identity/credentials/current.json, signs with its saved holder key, and returns only the presentation. It works without login or network access and does not modify the saved credential or key. Missing claims, expired credentials, mismatched keys, and unsupported disclosure formats fail instead of producing a presentation. It supports Link's flat SHA-256 disclosures; credentials with plaintext user claims or nested selective disclosures are rejected.
Send the returned presentation as the Identity-Presentation HTTP header.
The verifier still validates the issuer signature, holder signature, audience, nonce, expiry, and required claims. Presentations include a fresh signing time and should be sent promptly. Follow the service's retry instructions. Completed interactions may return a saved result; a new operation or expired interaction may require a fresh challenge. The Agent Identity SDK verifies nonce equality; the service manages interaction state and replay policy.
Local inspection uses the same MCP exclusion:
link-cli identity credentials list --format json link-cli identity attestations list --format json
These commands inspect local files without login or Link API calls and display metadata in both terminal and structured output. Credential inspection reports the saved ~/.link-cli/identity/credentials/current.json path, issuer, cached expiry/expired status, holder-key path/thumbprint, and claim names. Private keys are never opened; credentials, tokens, and claim values are never printed. Inspection does not modify files, verify signatures, or filter artifacts by the active account.
Attestation inspection reports paths, issuer/key identifiers, per-batch stored_token_count, aggregate total_token_count, and storage (pool or export) for JSON batches in ~/.link-cli/identity/attestations. Counts describe stored tokens; external usage is untracked and AATs have no embedded expiry. Empty stores return empty lists. Lists include per-file errors alongside valid entries.
A spend request moves through: create → request approval → approved (with credentials).
Required fields for a regular card create: merchant_name, merchant_url,
context, and amount. payment_method_id is optional — if omitted,
your default payment method will be used, or the first eligible one if no
default is set. Shared Payment Token requests instead require network_id;
Link Pay Token requests require credential_type=link_pay_token and the
DOM-derived merchant_account_id, and Link supplies their canonical merchant
identity. Card requests may also include --network-id to identify the merchant's
network profile.
Constraints: context must be at least 100 characters; amount is an integer in the currency's smallest unit (e.g. cents) and must not exceed 50000 USD cents equivalent; currency can be any supported 3-letter ISO code. The user has 30 minutes from when approval is requested to approve. Approved credentials (card or SPT) are valid for 48 hours from spend request creation.
Omitting currency uses the consumer's default spend-request currency, reported by
userinfo.default_spend_request_currency (user-info retrieve in the CLI).
Specify currency explicitly when using an amount from a checkout with a known
currency. Amounts, line-item unit_amount values, and total amount values use
the selected currency's minor units: 100 means ¥100 in JPY or $1 in USD.
The client passes amounts through unchanged and performs no FX conversion.
This default applies to newly created requests; existing requests retain their
stored currency.
For an application-owned return destination, supply --completion-callback-url
and --nonce together when creating a request. Both are optional; supplying only
one is rejected locally. The callback must be a valid URL of at most 2,048
characters. The nonce must be nonblank and at most 512 JavaScript/UTF-16 code
units. The API enforces live-mode HTTPS and URL security rules.
Generate an opaque, unpredictable nonce, save it unchanged, and verify an exact match when your application receives the callback:
NONCE="$(openssl rand -hex 32)" link-cli spend-request create \ --amount 5000 \ --currency usd \ --context "The user asked to purchase the selected item from Example Store for $50, including all shipping costs and taxes." \ --merchant-name "Example Store" \ --merchant-url "https://example.com" \ --completion-callback-url "https://agent.example.com/link/callback" \ --nonce "$NONCE"
Use the returned approval_url and status_details.requires_action.next_action.action_url
as-is. When creation-time callback context is supplied, do not manually append
callback parameters. A callback does not establish approval; retrieve the spend
request and check its status before continuing.
Test mode: Pass --test to create a testmode SpendRequest. A testmode SpendRequest will return test payment credentials (e.g test card 4000009990001984) rather than a real payment credential. Testmode SpendRequests will not charge the underlying payment method of the SpendRequest. This is useful for development and integration testing without real payment methods.
# Update before approval link-cli spend-request update lsrq_001 \ --merchant-url https://press.stripe.com/working-in-public # Request approval separately (alternative to create --request-approval) link-cli spend-request request-approval lsrq_001 # Retrieve a request; request a credential only when needed link-cli spend-request retrieve lsrq_001 # Cancel a spend request (from created, pending_approval, or approved state) link-cli spend-request cancel lsrq_001
Use spend-request retrieve <id> --interval 2 to wait for the initial status
to change. Polling starts only for created, pending_approval, or
requires_action with an auto_resume resolution. Other statuses, including
submitted and unfamiliar API values, return immediately. A change to another
waiting status also returns; inspect the result and retrieve again as needed.
If --timeout or --max-attempts is reached without a change, the command exits
non-zero with POLLING_TIMEOUT.
For a subscription or other recurring purchase, confirm the spending limit and
cadence with the user, then add --recurring-interval when creating the request:
link-cli spend-request create \ --credential-type card \ --merchant-name "Example Pro" \ --merchant-url "https://example.com/subscribe" \ --amount 2000 \ --currency usd \ --recurring-interval month \ --context "Monthly access to Example Pro requested by the user, with a spending limit of $20 per month including all taxes."
Supported intervals are day, week, month, and year.
--recurring-interval-count is a positive integer and defaults to 1 on the
server. For example, --recurring-interval week --recurring-interval-count 2
resets the limit every two weeks. The count requires an interval. Omit both
flags for a one-time purchase.
For recurring requests, --amount is the spending limit per period, in the
currency's minor units. It is not the total budget for the subscription's
lifetime. The merchant controls billing; the CLI does not schedule charges.
The same recurring options are available with card, shared_payment_token,
and link_pay_token; each credential type keeps its merchant targeting and
checkout requirements. Link enforces availability and applicable limits.
Approval and credential retrieval use the normal flow. Responses include
recurring.interval and recurring.interval_count; spend-request list also
shows the cadence. Recurring card requests remain active after individual
charges. Do not wait for the request to become succeeded after each payment
or create a new request for each period. Confirm each payment from the
merchant's result or transaction records.
Recurring terms are set at creation; spend-request update does not change
the interval or count. Respect the returned credential expiry.
To stop future use of the payment credential, run
link-cli spend-request cancel <id> and check the result. This revokes the
credential; it does not cancel the subscription with the merchant. Complete
the merchant's cancellation flow when the user asks to cancel the subscription.
By default, a spend request provisions a virtual card. Link can also provide a shared payment token (SPT) for use with the Machine Payment Protocol (MPP) or a Link Pay Token (LPT) for supported Stripe payment surfaces. LPT is a distinct credential type.
For merchants that support the Machine Payments Protocol (HTTP 402) and the Stripe payment method, instead pass --credential-type "shared_payment_token" when creating the spend request. SPTs for one-time requests are single use; a consumed token requires a new spend request.
For recurring SPTs, usage_limits.max_amount is the spending limit per period.
The available spending amount resets to this limit each new period, as defined
by recurring.interval and recurring.interval_count.
Use mpp decode to validate a raw WWW-Authenticate header and extract the network_id needed for shared_payment_token spend requests:
link-cli mpp decode \
--challenge 'Payment id="ch_001", realm="merchant.example", method="stripe", intent="charge", request="..."'Use mpp pay to complete purchases:
link-cli mpp pay https://climate.stripe.dev/api/contribute \ --spend-request-id lsrq_001 \ --method POST \ --data '{"amount":100}' \ --header "X-Custom: value"
Some Stripe payment surfaces expose an AI-agent steering block that supports a
Link Pay Token (LPT). Inspect the checkout in a browser before creating the
SpendRequest: find the .AiAgentPaymentSteering block and verify that both
input[name="link_pay_token"] and
data-stripe-merchant-account="acct_..." are present in the same Stripe
frame.
Create an LPT-bound request with the DOM-derived account ID. Do not pass
--merchant-name or --merchant-url; Link resolves the canonical merchant
identity from the account ID for the approval screen.
link-cli spend-request create \
--credential-type link_pay_token \
--merchant-account-id acct_... \
--payment-method-id csmrpd_xxx \
--context "Purchasing an item from the checkout the agent inspected. The user initiated this purchase through the shopping assistant." \
--amount 3500 \
--request-approvalLPT requests do not support --test, --network-id, merchant name, or merchant
URL. After approval, retrieve --include link_pay_token immediately before
using it on the same checkout surface. Each returned LPT is valid for up to 30
minutes, or until the SpendRequest expires. If either DOM marker is absent,
create a regular virtual card SpendRequest instead; do not create an LPT
request.
| Limit | Value |
|---|---|
| Max amount per spend request | $500 (50,000 cents) |
| Approval window | 30 minutes — user must approve within 30 min of request-approval |
| Card / SPT validity | 48 hours from spend request creation |
| Daily spend | $500 |
| Monthly spend (30 days) | $20,000 |
| Concurrent active requests (created + approved) | 30 |
| Concurrent approved requests | 10 |
| Hourly creation rate | 50 per hour |
| Rolling creation rate | 200 per 60 days |
Dollar limits are enforced as the USD equivalent for spend requests in other currencies.
Use mpp pay to complete purchases on merchants that use the Machine Payments Protocol. The spend request must use credential_type: "shared_payment_token" and you must approve it before paying. Requests created by mpp pay are for one-time payments; consumed SPTs require a new spend request.
link-cli mpp pay https://climate.stripe.d
| |
| |
| |
| |
| |
| |
| |