Back to MCP Servers

Plaid MCP Server

Official Plaid Dashboard MCP server for diagnosing Plaid Items, analyzing Link conversion funnels, and inspecting integration health and API usage.

Payments by Plaid OAuth2 active
Overview

The Plaid Dashboard MCP server is an official, Plaid-hosted Model Context Protocol endpoint that lets LLM-powered applications connect to Plaid's developer tools. It is designed to help engineers and support teams understand integration health and investigate user-facing issues without manually navigating the Plaid Dashboard. The server is currently described by Plaid as under active development with limited official support.

The server exposes a small set of read-only tools for diagnosing Plaid Items, analyzing Link conversion funnels, retrieving product usage and API request volumes, and listing teams associated with the authenticated account. Because it operates on production data, it requires a production Plaid account that has been approved for at least one product. Authentication uses OAuth 2.0 client credentials scoped to mcp:dashboard, with short-lived 15 minute access tokens and a refresh token flow.

The endpoint is served over the MCP Streamable HTTP transport at https://api.dashboard.plaid.com/mcp, and is accessed by passing an Authorization: Bearer <access_token> header. Plaid's docs include example tool configurations for the OpenAI and Anthropic client libraries.

Tools

Tool Description
plaid_debug_item Diagnose a Plaid Item by retrieving related metadata to investigate why it may not be working properly.
plaid_get_link_analytics Analyze Plaid Link conversion funnels, user progression through Link, and error rates.
plaid_get_usages Retrieve product usage metrics and API request volumes for the account.
plaid_list_teams List the teams associated with the OAuth token's account.
Setup Guide

Prerequisites

  • A Plaid account with production access approved for at least one product
  • Production client_id and secret

Step 1: Mint an OAuth access token

Request a client_credentials token scoped to mcp:dashboard:

curl -X POST https://production.plaid.com/oauth/token \
  -H 'Content-Type: application/json' \
  -d '{
    "client_id": "YOUR_PLAID_CLIENT_ID",
    "client_secret": "YOUR_PRODUCTION_SECRET",
    "grant_type": "client_credentials",
    "scope": "mcp:dashboard"
  }'

The response contains an access_token (15 minute TTL) and a refresh_token. When the access token expires, exchange the refresh token for a new one:

curl -X POST https://production.plaid.com/oauth/token \
  -H 'Content-Type: application/json' \
  -d '{
    "client_id": "YOUR_PLAID_CLIENT_ID",
    "secret": "YOUR_PRODUCTION_SECRET",
    "refresh_token": "YOUR_REFRESH_TOKEN",
    "grant_type": "refresh_token"
  }'

Step 2: Configure your MCP client

The server is hosted by Plaid at https://api.dashboard.plaid.com/mcp and uses the MCP Streamable HTTP transport. Pass the access token as a Bearer header.

OpenAI tool configuration (from Plaid docs):

{
  "type": "mcp",
  "server_label": "plaid",
  "server_url": "https://api.dashboard.plaid.com/mcp",
  "require_approval": "never",
  "headers": { "Authorization": "Bearer <dashboard_token>" }
}

For MCP clients that support remote HTTP servers with custom headers, point them at the same URL and inject the Authorization: Bearer <access_token> header. Expired tokens return HTTP 401; refresh or mint a new token to recover.

Use Cases
  • Diagnose a misbehaving Plaid Item end-to-end by asking an agent to pull metadata and surface likely causes of failure
  • Triage a drop in Link conversion by asking the agent to break down funnel progression and error rates over the last 7 days
  • Monitor monthly product usage and API request volumes across environments to forecast billing and capacity
  • Build an internal Slack or chat bot that resolves customer support tickets by inspecting Items referenced by support engineers
  • Audit which teams in your Plaid account exist and route troubleshooting questions to the right one
Example Prompts
  • "Debug Plaid Item item_abc123 and tell me why it might be failing."
  • "Show me Link conversion funnel metrics for the last 30 days and call out any steps with elevated error rates."
  • "What were our Auth and Transactions API request volumes last month?"
  • "List all teams on our Plaid account."
  • "Compare this week's Link drop-off rate to last week's and summarize the biggest regressions."
Pros
  • Official, Plaid-hosted MCP server, no self-hosting or community fork to maintain
  • Direct access to Dashboard-only data such as Link analytics, usage metrics, and Item debug metadata
  • Uses standard OAuth 2.0 client credentials with refresh tokens, easy to wire into existing auth infrastructure
  • Streamable HTTP transport works with mainstream MCP clients and the OpenAI/Anthropic tool APIs
Limitations
  • Marked as under active development with limited official support
  • Production access approval for at least one Plaid product is required, no sandbox usage
  • Short 15 minute access token lifetime requires implementing the refresh flow
  • Read-only diagnostic surface, no tools for executing Plaid product API calls like Transactions sync or Auth
Alternatives
  • Self-built MCP wrapper around the Plaid REST API for non-Dashboard operations (Transactions, Auth, Identity)
  • Stripe's official MCP server, for teams whose payments stack centers on Stripe rather than Plaid
  • Community Plaid MCP implementations on GitHub (for example argotdev/plaid-mcp-ts), which expose Plaid product APIs but are not maintained by Plaid