Read bank data through Plaid — accounts, balances, transactions, identity and liabilities — and manage the Items that connect them.
Plaid ships in the w6w first-party pack. It declares 14 actions, 2 health checks, and the host runs its code in a sandbox that never sees the credential.
io.w6w.plaidPlaid brings bank data into a workflow — account details, balances, transactions, identity information and liabilities — and manages the Items that connect a person’s bank to an application. Reading transactions uses Plaid’s sync model rather than a fixed date range: each call returns exactly what changed since the last cursor, so a workflow stays correct as pending charges settle, amounts get corrected and transactions occasionally get removed, without re-reading or double-counting anything.
Account actions distinguish a cached balance, fast and free of the bank’s own rate limits, from a live balance fetched fresh from the institution for anything approaching a payment decision. Identity and liabilities actions return what a bank itself believes about an account holder and their loans or cards — evidence rather than self-reported data. Connecting a new bank still requires a human in Plaid’s own Link flow, but this app handles both server-side halves: minting the token Link needs and exchanging what it returns for a working connection.
Item management actions read a connection’s health (including when a user’s bank login has expired and needs re-authentication), point a connection’s webhook at a new URL, and disconnect a bank entirely when access should end.
Three routes to the same 14 actions. The Workflow tab is generated from Plaid's own manifest and carries its real ids, so it is copy-pasteable; the Code and CLI examples are the same call for any action on any app, so every app-specific value in them is a blank you fill in.
account-list The Item's accounts with CACHED balances — fast and possibly hours stale. Use Get Balance when the number has to be current.
auth-get The bank details an ACH debit needs — the most sensitive call here. Read it when setting up a payment, store only what the processor needs, and never log it.
balance-get Balances fetched from the bank rather than Plaid's cache — slower, rate limited by the institution, and the only version safe to make a payment decision on.
identity-get Names, addresses, emails and phones as the BANK holds them — which is what makes a name match evidence. A joint account has several owners.
institution-list Supported banks, with the status Plaid publishes for each — a far better answer to 'why did this fail' than a generic error.
item-get One bank connection's state — including the error that says its credentials have expired, which is otherwise indistinguishable from a quiet account.
item-remove Disconnect a bank account permanently — invalidating its token, ending billing, and ending Plaid's access. Reconnecting produces a NEW Item with a new id.
liabilities-get Credit cards, student loans and mortgages with their APRs, minimum payments and due dates — the terms, not just the balance. Coverage varies by institution.
link-token-create Mint the short-lived token Plaid Link needs in a browser. With an access token it opens in update mode, which is how an expired Item is repaired without losing its history.
public-token-exchange Turn the browser's short-lived public token into an Item's access token. That token is a long-lived credential for somebody's bank data — store it like a password.
sandbox-item-create Make an Item with no browser — sandbox only. This is what makes a Plaid workflow testable end to end, since every other environment needs a human in Plaid Link.
transaction-refresh Force an immediate fetch from the bank. Billable on most plans, and it returns an acknowledgement rather than data — the transactions arrive asynchronously.
transaction-sync What has changed since a cursor — added, modified and removed, separately. The correct way to follow an account, because bank transactions are amended after the fact.
webhook-update Point one Item's notifications at a URL. The alternative to polling — Plaid refreshes a few times a day, so most polls cost a call and return nothing.
A workflow step names the app and the action, and the editor fills in the
connection when you pick one. This is the Step shape from the
workflow spec, carrying Plaid's real ids.
{
"manifestVersion": "2",
"name": "plaid-example",
"steps": [
{
"id": "account-list",
"uses": {
"app": "io.w6w.plaid",
"action": "account-list",
"connection": "conn_YOUR_CONNECTION_ID"
},
"with": {
"accessToken": "<accessToken>"
}
}
]
}account-list auth-get balance-get identity-get institution-list +9 more actions available
Every app-specific value here is a blank you have to fill in. An
app action is reached through the connection that authenticates it, so the
address is a connection id, not the app id — and connections belong to your account,
so a public page cannot know yours. Create one for Plaid, then fill in
the three blanks: conn_YOUR_CONNECTION_ID, the action key, and the
parameters that action declares. The call itself is real — the shape is transcribed
from the studio's own snippet builder, which prints the same kind of blanks — but
nothing in it is specific to Plaid. The Workflow tab is where this app's
real ids are.
npm install @w6w/sdkyarn add @w6w/sdkpnpm add @w6w/sdkdeno add npm:@w6w/sdkimport { W6wClient, isActionRun } from "@w6w/sdk";
// Reads W6W_BASE_URL and W6W_TOKEN from the environment when omitted.
const client = new W6wClient();
const envelope = await client.run({
urn: "conn_YOUR_CONNECTION_ID",
action: "account-list",
payload: {
accessToken: "<value>",
// accountIds: "<value>",
},
});
if (isActionRun(envelope)) console.log(envelope.value); npm install -g @w6w/cli w6w run conn_YOUR_CONNECTION_ID --action account-list --payload '{"accessToken":"<value>"}' Give an AI agent Plaid — without giving it Plaid's credentials. One MCP endpoint exposes every app, function and workflow the caller is entitled to, as tools it can discover and run. Access is granted per team while we onboard.
One tool call{
"name": "w6w_invoke",
"arguments": {
"ref": "app:io.w6w.plaid#account-list",
"input": {
"accessToken": "<accessToken>"
}
}
}
Every tool names its target with a single ref. The
app: form above doesn't name a connection at all — the
host resolves which of the caller's Plaid connections to sign
with, and refuses rather than guesses when the answer is ambiguous.
The token is attached host-side, at the moment of the call. It is never a tool argument, never in the model's context, and never in a transcript — so a prompt injection has nothing to exfiltrate.
Tools are derived per end user from what that person has actually connected and is entitled to — not one shared bot identity carrying the union of everyone's access.
Multi-step work runs on the workflow engine and returns a run handle the agent can poll — retries, branching and state survive the conversation that started them.
Plaid's declared health checks are on the surface too, so an agent can tell "the vendor is down" from "your credential expired" before it burns a retry on either.
The MCP surface is part of the hosted platform. Plaid itself is MIT, and the runtime that executes it is source-available (FSL).
Plaid declares its own checks, so its health is a property of the app rather than something the host guesses at.
Plaid's API and Link components. Institution-level trouble is per Item and shows up in `item-get`'s error rather than here.
Probes an Item-free endpoint, so a failure can only mean the connection is wrong — most often a secret from the other environment. Per-Item errors are a different question.