Connect an AI Coding Agent

This tutorial sets up an AI coding agent to interact with your PlaidCloud workspace using the Model Context Protocol (MCP). Once configured, you can ask the agent to read tables, run workflows, build allocations, and answer questions about your data — directly from your editor.

Takes about 20 minutes. Works with Claude Code, Cursor, Claude Desktop, ChatGPT, GitHub Copilot, and any MCP-compatible client.

What You’ll Build

A working connection between your AI coding agent and your PlaidCloud workspace, where the agent can:

• List projects, tables, workflows, and dimensions
• Read table contents
• Inspect workflow definitions
• Trigger workflow runs
• Answer questions about your data without you switching contexts

Prerequisites

• A PlaidCloud workspace (start a free trial if you don’t have one)
• An AI coding agent you already use — Claude Code, Cursor, Claude Desktop, ChatGPT, Copilot, or Gemini
• The agent must support MCP (most current AI tools do)

Step 1: Find Your Workspace’s MCP URL

Every PlaidCloud workspace exposes an MCP endpoint at:

https://<your-workspace>.plaid.cloud/mcp/

Replace <your-workspace> with your workspace subdomain — the same one you use to sign in to the PlaidCloud UI.

Step 2: Get an Authentication Token

1. While signed in to PlaidCloud in a browser, visit:

   https://<your-workspace>.plaid.cloud/mcp/setup/token

2. Copy the bearer token shown on the page. Keep it safe — it grants the same access your account has.

Caution

Treat the token like a password. It bypasses interactive authentication and acts on your behalf. If it leaks, revoke it from the same endpoint.

Step 3: Configure Your Agent

Claude Code

Run in your terminal:

claude mcp add --transport http plaidcloud https://<your-workspace>.plaid.cloud/mcp/

For static-token authentication (no OAuth flow, simpler for long sessions), open the URL from Step 2 in a browser, copy the displayed config snippet, and paste it into your .mcp.json file.

See Claude Code setup for full options.

Cursor

Get a Bearer token from https://<your-workspace>.plaid.cloud/mcp/setup/token and add this to your Cursor MCP config:

{
  "mcpServers": {
    "plaidcloud": {
      "url": "https://<your-workspace>.plaid.cloud/mcp/",
      "headers": {
        "Authorization": "Bearer <token>"
      }
    }
  }
}

See Cursor setup for full options.

Claude Desktop

Open Settings → Developer → MCP Servers and add:

• Server URL: https://<your-workspace>.plaid.cloud/mcp/
• Use OAuth login when prompted

See Claude Desktop setup for full options.

ChatGPT

1. Settings → Connectors → Add custom connector
2. Enter:
   • Name: PlaidCloud
   • MCP server URL: https://<your-workspace>.plaid.cloud/mcp/
3. ChatGPT redirects you to PlaidCloud for OAuth login. Approve the connection.
4. Toggle the connector on inside any conversation that should use it.

See ChatGPT setup for full options.

Copilot

Get a Bearer token and add to .vscode/mcp.json:

{
  "servers": {
    "plaidcloud": {
      "url": "https://<your-workspace>.plaid.cloud/mcp/",
      "headers": {
        "Authorization": "Bearer <token>"
      }
    }
  }
}

VSCode reads this on startup and on file change.

See Copilot setup for full options.

Step 4: Verify the Connection

Ask the agent something simple:

“List the projects in my PlaidCloud workspace.”

The agent should respond with your project list. If it doesn’t, check the troubleshooting steps below.

Step 5: Try Real Tasks

Once connected, try:

• Explore your data: “What columns does the sales table in my Quickstart project have?”
• Run a workflow: “Trigger the monthly_close workflow in the Financials project and let me know when it finishes.”
• Build something: “Create a Table Lookup step in my Test workflow that filters orders to the last 30 days.”

The agent’s responses will be grounded in your actual workspace state — not generic answers.

Troubleshooting

If the agent can’t reach PlaidCloud or returns auth errors:

• Token expired — refresh it at https://<your-workspace>.plaid.cloud/mcp/setup/token
• Wrong scopes — some tools require specific PlaidCloud scopes (e.g., analyze.workflow.write). Run mcp_introspect(name='<tool>') in the agent to see required scopes
• Workspace subdomain wrong — confirm by signing into the PlaidCloud UI; the subdomain is the part before .plaid.cloud

See AI coding agents troubleshooting for more.

What’s Next

• AI coding agents getting started — base setup details
• AI coding agents reference — every supported agent and its setup
• Concepts — the PlaidCloud data model the agent will be reasoning over