---
title: "Plaid MCP Servers: review, setup & limitations · Undominated.ai"
canonical: https://undominated.ai/mcp-servers/plaid/
description: "Plaid publishes two official MCP surfaces. The local sandbox server in the plaid/ai-coding-toolkit repository generates mock financial data, searches Plaid documentation, mints sandbox access tokens and simulates webhooks — deliberately scoped so an agent cannot touch live financial data. The hosted Dashboard MCP at api.dashboard.plaid.com serves Production diagnostics — debugging Items, Link conversion data and usage metrics — over Streamable HTTP with an OAuth token scoped to mcp:dashboard."
---

# Plaid MCP Servers: review, setup & limitations · Undominated.ai

> Plaid publishes two official MCP surfaces. The local sandbox server in the plaid/ai-coding-toolkit repository generates mock financial data, searches Plaid documentation, mints sandbox access tokens and simulates webhooks — deliberately scoped so an agent cannot touch live financial data. The hosted Dashboard MCP at api.dashboard.plaid.com serves Production diagnostics — debugging Items, Link conversion data and usage metrics — over Streamable HTTP with an OAuth token scoped to mcp:dashboard.

[← Explore all mcp servers](/mcp-servers/)

FINANCIAL DATA / Plaid

# Plaid MCP Servers

Plaid publishes two official MCP surfaces. The local sandbox server in the plaid/ai-coding-toolkit repository generates mock financial data, searches Plaid documentation, mints sandbox access tokens and simulates webhooks — deliberately scoped so an agent cannot touch live financial data. The hosted Dashboard MCP at api.dashboard.plaid.com serves Production diagnostics — debugging Items, Link conversion data and usage metrics — over Streamable HTTP with an OAuth token scoped to mcp:dashboard.

 [See setup guidance ↓](#setup)[Original source ↗](https://plaid.com/docs/resources/mcp)

SOURCE REVIEW

 Reviewed 2026-10-07
 Evidence 4 linked sources
 Publisher Plaid
 Licence [MIT ↗](https://github.com/plaid/ai-coding-toolkit/blob/c465460b1ce203659887fd5daaa0297f4e275003/LICENSE)
 Revision c465460b1ce2
 [Read what was—and wasn’t—checked ↓](#review)

“Model Context Protocol (MCP) server to build fast integration testing with Plaid Sandbox data.”

 [Plaid · upstream description ↗](https://github.com/plaid/ai-coding-toolkit/blob/c465460b1ce203659887fd5daaa0297f4e275003/sandbox/pyproject.toml) Our analysis follows below.

01 / THE REASONING

## Why this made the selection.

 - Brief-named official finance gap: Plaid documents both a local sandbox MCP server and a hosted Dashboard MCP on its own docs site, with the local implementation maintained in the plaid organisation under MIT.
- The split is safety-shaped: the local server is sandbox-only by design, while Production diagnostics require a separate OAuth token with the mcp:dashboard scope and Production access approval.

### A good fit for

 - Building a Plaid integration with an agent using sandbox mock data, docs search and webhook simulation before touching Production.
- Production debugging on the hosted Dashboard MCP — Item diagnostics, Link conversion and usage metrics.

### Weigh up before choosing

 - The Dashboard MCP works with Production data only, requires Production access approval, and Plaid states it is under active development with limited support and may see breaking changes.
- The local server is sandbox-only by design and cannot exercise live financial data.

02 / THE REVIEW RECORD

## What we actually inspected.

Source review has boundaries. A clear record is more useful than a “safe” badge.

### Material inspected

 - Repository README at pinned revision c465460b1ce203659887fd5daaa0297f4e275003: the toolkit provides local MCP tools (mock data generation, documentation search) and product-specific guides; /sandbox holds the sandbox MCP server; links to the Plaid Dashboard MCP and Plaid CLI.
- sandbox/pyproject.toml at the same revision: package mcp-server-plaid version 0.1.1, MIT licence, requires Python >= 3.11 with mcp>=1.6,<2 and plaid-python, entry point mcp-server-plaid = mcp_server_plaid:main.
- sandbox/README.md at the same revision: four tools — search_documentation, get_mock_data_prompt, get_sandbox_access_token, simulate_webhook — with a uvx configuration that takes sandbox client-id and secret; LICENSE at the revision is MIT (Copyright 2025 Plaid Inc.).
- Plaid docs MCP page: the Dashboard MCP server is hosted at https://api.dashboard.plaid.com/mcp/ over Streamable HTTP, requires an OAuth token with the mcp:dashboard scope created with the client_credentials grant, works with Production data only, and is under active development with limited support.

### Our findings

 - The two surfaces are complementary: local sandbox tooling for integration development, hosted diagnostics for Production operations.
- The local server's credentials are sandbox keys by explicit README instruction.

### Not established by this review

 - The package was not installed and no Plaid account was connected.

The review applies to the material and revision named here. A newer upstream release can change its behavior.

03 / PUT IT TO WORK

## Connect a server deliberately.

[Upstream setup instructions ↗](https://github.com/plaid/ai-coding-toolkit/blob/c465460b1ce203659887fd5daaa0297f4e275003/sandbox/README.md)
 - Obtain sandbox credentials from the Plaid Developer Dashboard under Developers → Keys.
- Add the configExample to the MCP client, replacing the placeholders with sandbox credentials.
- For the hosted Dashboard MCP, create an OAuth token with the mcp:dashboard scope and point a remote-MCP client at https://api.dashboard.plaid.com/mcp/.

### Before you start

 - A Plaid developer account with sandbox credentials for the local server.
- Production access approval for the Dashboard MCP.

### Documented configuration example

Use the client and authentication context described in the setup guidance. Configuration formats vary between hosts.

 {
 "mcpServers": {
 "plaid": {
 "command": "uvx",
 "args": [
 "mcp-server-plaid",
 "--client-id",
 "YOUR_PLAID_CLIENT_ID",
 "--secret",
 "YOUR_PLAID_SECRET"
 ]
 }
 }
} Copy configuration ↗

### Compatibility

Claude, Cursor, Codex, VS Code and Zed can run the local server as a subprocess via uvx or pip. · Claude.ai, ChatGPT and other remote-MCP clients can connect to the Dashboard MCP over Streamable HTTP.

### Implementation

Python

### Transports

stdio · Streamable HTTP

### Plaid sandbox credentials locally; Production OAuth token for the Dashboard MCP

 - The local server holds sandbox client credentials and touches only sandbox data.
- The Dashboard MCP acts within the mcp:dashboard scope on Production diagnostics.

### Cost model

No separate MCP charge is documented; standard Plaid plan and usage terms apply.

04 / FOLLOW THE EVIDENCE

## The source trail.

Our notes are separate from the original resource. Check upstream before adopting a new version.

 - [Plaid AI coding toolkit README at pinned revision ↗](https://github.com/plaid/ai-coding-toolkit/blob/c465460b1ce203659887fd5daaa0297f4e275003/README.md) Checked 2026-10-07 https://github.com/plaid/ai-coding-toolkit/blob/c465460b1ce203659887fd5daaa0297f4e275003/README.md Supports: summary, whySelected, repository, revision
- [Plaid sandbox MCP server manifest at pinned revision ↗](https://github.com/plaid/ai-coding-toolkit/blob/c465460b1ce203659887fd5daaa0297f4e275003/sandbox/pyproject.toml) Checked 2026-10-07 https://github.com/plaid/ai-coding-toolkit/blob/c465460b1ce203659887fd5daaa0297f4e275003/sandbox/pyproject.toml Supports: language, transports, upstreamDescription, descriptionSourceUrl
- [Plaid sandbox MCP server README at pinned revision ↗](https://github.com/plaid/ai-coding-toolkit/blob/c465460b1ce203659887fd5daaa0297f4e275003/sandbox/README.md) Checked 2026-10-07 https://github.com/plaid/ai-coding-toolkit/blob/c465460b1ce203659887fd5daaa0297f4e275003/sandbox/README.md Supports: install, configExample, compatibility, bestFor
- [Plaid docs: MCP server resources ↗](https://plaid.com/docs/resources/mcp) Checked 2026-10-07 https://plaid.com/docs/resources/mcp Supports: homepage, access, limitations, transports

 [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/)

## Continue your investigation

 - [Inspect resource evidence](/tools/)
- [Choose a host](/agents/)
- [Inspect reusable workflows](/skills/)
- [Use Undominated data](/api/)
