MCP server for Pluggy — Open Finance Brasil aggregator (ITP/TPP). Exposes Brazilian banking APIs for account discovery, transactions, balances, identities, and payments initiation (PISP).
pluggy-mcp demonstrates solid fundamentals: all 18 tools have clear verb-noun naming (list_, get_, create_, update_, delete_), descriptions are present and contextual (avg ~120 chars), and input schemas use Zod with type definitions and descriptions. However, output schemas are entirely undocumented, the code shows tools return `ok(await pluggyRequest(...))` with no schema definition visible. Error handling is minimal (generic `ok()` wrapper). Parameter descriptions lack format constraints (e.g., 'ISO date' but no regex/pattern). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. Pagination is supported (page, pageSize params) but no documented limit caps or total counts in responses. Sensitive operations (delete_item, create_item with credentials) lack confirmation/dry-run patterns.
Mint a connect token for embedding the Pluggy Connect widget on the client. Pluggy endpoint: POST /connect_token.
Create a new bank connection (item) for a connector. Pluggy endpoint: POST /items.
Create a payment intent for Pluggy Payments (PISP). Pluggy endpoint: POST /payments/intents.
Delete a bank connection (revokes credentials, removes accounts/transactions). Pluggy endpoint: DELETE /items/{id}.
Get a single account by id. Pluggy endpoint: GET /accounts/{id}.
Get a single connector definition by id. Pluggy endpoint: GET /connectors/{id}.
Output schemas completely undocumented. Tools return raw API responses via `ok(await pluggyRequest(...))` with no schema definition visible in code. LLMs cannot predict response structure, field types, or required fields for downstream tool chaining.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. delete_item, create_item, update_item, create_payment_intent are destructive/write operations but lack annotations to guide agent safety.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
Fetch a single investment by id. Pluggy endpoint: GET /investments/{id}.
Fetch a single bank connection by id. Pluggy endpoint: GET /items/{id}.
Fetch the current status of a payment intent. Pluggy endpoint: GET /payments/intents/{id}.
Get a single transaction by id. Pluggy endpoint: GET /transactions/{id}.
List accounts (checking, savings, credit card, investment) tied to an item. Pluggy endpoint: GET /accounts?itemId=...
List Pluggy's transaction categorization taxonomy. Pluggy endpoint: GET /categories.
Lists supported Brazilian banks (connectors). Pluggy endpoint: GET /connectors. Optional filters: name, types (e.g. PERSONAL_BANK, BUSINESS_BANK), countries (BR), sandbox.
Fetch identity data (legal name, document, address) for an item. Pluggy endpoint: GET /identity?itemId=...
List transactions for a specific investment. Pluggy endpoint: GET /investments/{id}/transactions.
List investments collected for an item. Pluggy endpoint: GET /investments?itemId=...
List transactions for an account in a date range. Pluggy endpoint: GET /transactions?accountId=&from=&to=.
Refresh / update credentials for an existing bank connection. Pluggy endpoint: PATCH /items/{id}.
Sensitive credential parameters (e.g., 'parameters' object in create_item containing CPF/password) lack validation, sanitization, or confirmation. No dry-run or confirmation-request pattern for destructive operations.
Parameter descriptions lack format constraints. 'ISO date lower bound (yyyy-mm-dd)' is stated but no regex pattern or format validator visible. Date parameters (from, to) could accept invalid formats without validation.
Pagination support (page, pageSize) present but no documented result limits or total count fields. Large result sets could exhaust context window. No guidance on max pageSize or default behavior.