Trello MCP in Cursor and VS Code

To use Trello in Cursor, add Trello’s official MCP server, which Atlassian hosts at https://mcp.trello.com/v1. Atlassian publishes a one-click install link for Cursor; the manual way is a url entry in .cursor/mcp.json. Either way Cursor opens Atlassian’s sign-in, you choose one Trello workspace and the permissions to grant, and the agent can then read, search, create, move and archive cards with your own Trello rights. VS Code and Codex take the same URL. Trello’s server is not the Atlassian Rovo MCP Server that covers Jira and Confluence; it is a separate server, and it signs in with OAuth only. Because Trello names its tools by read and write, it is also one of the easiest servers to limit to reads.

This guide covers the editors and Codex. The terminal setup with Claude Code is in Trello MCP with Claude Code, and whether an agent should use MCP, a CLI or the REST API at all is in Trello MCP or CLI. Neither is repeated here.

Which Trello server, today

Atlassian’s Trello MCP documentation↗ (opens in a new tab) names Cursor, Visual Studio Code and Gemini CLI among the supported clients, and gives one address for adding Trello by hand. The facts that shape the setup:

• Trello data only. The Rovo MCP Server covers Jira, Confluence, Compass, Bitbucket and more; Atlassian says Trello data may come to it in a future release, but it is not there now. If you use both, you add two servers.
• OAuth 2.0 only, through the browser. There is no token you can put in a header instead.
• One workspace per connection, chosen on the consent screen, with read, write and search granted separately. Multi-workspace support is planned.
• No permanent delete. The assistant can archive cards and lists, which you can restore, but not delete them.
• Any Trello plan. Creating focus time in Planner is the exception and needs Premium or Enterprise.

Because this server is official, there is little reason to install a community Trello server in an editor. If you do, it runs with an API key and token that can reach your whole account; read MCP security risks first.

Cursor

Atlassian’s repository for the server↗ (opens in a new tab) links a Cursor deeplink that opens Cursor and installs the server for you. By hand, add this to ~/.cursor/mcp.json for every project or .cursor/mcp.json for one:

{
  "mcpServers": {
    "trello": {
      "url": "https://mcp.trello.com/v1"
    }
  }
}

1. Open Customize in Cursor’s sidebar, find trello under MCPs and switch it on.
2. Atlassian’s sign-in opens in your browser. Sign in with the account you use for Trello.
3. On the consent screen, pick the workspace and review read, write and search, plus the separate Inbox and Planner permissions. Untick write if you want to start with reads only.
4. Back in Cursor the server shows its tools. If it does not, the Output panel’s MCP Logs show why.

The file holds only the address, so a project copy can be committed and each person signs in as themselves. If you left a permission out and need it later, Atlassian’s advice is to disconnect and connect again with the wider set.

Let reads run, keep writes on ask

Trello’s tools are grouped by object and split by intent: trelloReadMember, trelloReadBoard, trelloReadList, trelloReadCard, trelloReadChecklist, trelloReadInbox, trelloReadPlanner and trelloSearch read, and the matching trelloWrite… tools change things. That makes Cursor’s allowlist easy. Cursor’s permissions reference↗ (opens in a new tab) takes server:tool entries in permissions.json, with * allowed inside a name:

{
  "mcpAllowlist": [
    "trello:trelloRead*",
    "trello:trelloSearch"
  ]
}

With the run mode set to Allowlist under Settings, Agents, Approvals & Execution, reads run straight away and every trelloWrite… call asks, with its arguments shown. Never allow trello:*, and do not use Run Everything with a board other people work from. A list in the file replaces the one in Cursor’s settings for MCP, so keep every entry in the file.

VS Code

Add the server to .vscode/mcp.json to share it with the repository, or run “MCP: Add Server” and choose HTTP to put it in your user profile. The workspace file uses servers, not mcpServers:

{
  "servers": {
    "trello": {
      "type": "http",
      "url": "https://mcp.trello.com/v1"
    }
  }
}

Start the server from the inline action above the entry and complete the sign-in. Then pick the right session. VS Code’s harness documentation↗ (opens in a new tab) says Copilot sessions can currently reach only local MCP servers that do not require authentication, and Trello is remote with a sign-in, so it will not appear there. Use a Local session. When a tool runs, VS Code lets you approve it once, for the session, the workspace or always; give the longer approvals to the read tools only. File details are in VS Code mcp.json, and approvals and trust in VS Code MCP security.

Codex

Atlassian does not document Codex for Trello, but the server is a standard remote server with OAuth, which OpenAI’s Codex MCP documentation↗ (opens in a new tab) covers. Add it and sign in:

codex mcp add trello --url https://mcp.trello.com/v1
codex mcp login trello

Codex keeps the entry in ~/.codex/config.toml, shared with its IDE extension and the ChatGPT desktop app. To make a connection that can only read, list the read tools in enabled_tools; the write tools are then hidden from the model altogether, whatever you granted on the consent screen:

[mcp_servers.trello]
url = "https://mcp.trello.com/v1"
enabled_tools = [
  "trelloReadMember",
  "trelloReadBoard",
  "trelloReadList",
  "trelloReadCard",
  "trelloReadChecklist",
  "trelloSearch",
]
default_tools_approval_mode = "prompt"

When you want writes back, remove enabled_tools and keep default_tools_approval_mode = "prompt", so Codex asks before each call. Codex also has a writes mode that asks only for tools the server does not mark as read-only; check what the server declares before you rely on it. /mcp in a session shows whether the server is connected.

Prompts that suit an editor

• “Which Trello workspace are you connected to, and what boards are in it?” Confirms the connection before anything else.
• “Create a card in To Do on the Sprint board for each TODO comment in src/billing, with the file and line in the description.”
• “Move ‘Fix login bug’ to Done and add a checklist item ‘Deployed’ ticked.”
• “Archive the cards in Done on the Sprint board that were finished before this week.” Archive, not delete, so it can be undone.

Ask for one board or one list at a time. The tools return card previews and the agent has to page through for the rest, so “how many cards” is more reliable than “list every card”.

Limits in an editor

• No comments, attachments, custom fields or card history yet. Atlassian lists all of them as coming, so a coding agent cannot yet leave a comment on the card it just moved.
• No machine sign-in. With OAuth only, a CI job cannot use this server; that case belongs to the REST API, as Trello MCP or CLI explains.
• Organisation controls. Trello MCP uses the same admin framework as the Rovo MCP Server, so on Enterprise an admin can block client domains or switch off read or write access. If sign-in is refused at work, that is where to ask.
• It acts as you. Changes appear in Trello under your name, not the agent’s.

The same editors, a board that names the agent

Trello gives you any number of lists, and its server works through your own account. fenbs connects to Cursor, VS Code and Codex the same way, with https://fenbs.ai/api/mcp and a browser sign-in, and adds what Trello’s server does not: commenting is one of its tools, a hand-issued token under Settings (name, scopes, optional expiry) covers the CI case, and every change is recorded under the assistant’s name. It has four fixed lanes, To Do, Next Up, In Progress and Completed, no due dates, no Power-Ups and no Trello importer; you paste a list of tasks instead. fenbs sets no tool annotations yet, so Codex’s writes mode asks for every fenbs tool. See connecting Cursor and fenbs vs Trello.