Trello MCP server

What it does

This server lets an agent read and edit Trello boards: lists, cards, comments, checklists, labels, members, attachments and custom fields. It is @delorenj/mcp-server-trello, a community package on npm, and Atlassian does not maintain it. Trello has no MCP server of its own. Statio runs version 1.8.1 with your API key and token, and the token acts as you, on every board you can open.

Tools

Version 1.8.1 lists 57 tools. None of them carry read-only or destructive hints, so the last column comes from what each tool calls in the source. Many take an optional boardId. Without one they use the active board, which set_active_board picks and the server saves to ~/.trello-mcp/config.json so it survives a restart.

Boards and workspaces

Lists

Cards

Comments

Checklists

Members and labels

Attachments and custom fields

Health

Set it up

Both setups need an API key and a token. Trello's authorization guide gets the key from the Power-Up admin page at trello.com/apps/admin: open or create a Power-Up, go to the Trello Auth tab and generate a key. The token comes from an authorize URL with that key in it:

https://trello.com/1/authorize?expiration=30days&scope=read&name=Trello%20MCP&response_type=token&key=YOUR_API_KEY

Click Allow and Trello shows the token. Two parameters decide most of what an agent can do. scope is read, write or both (read,write), and a read token makes Trello refuse every write tool above whatever your client allows. expiration is 1hour, 1day, 30days or never. Leave out account, which lets the token read your email and edit your profile, since no tool here needs it.

Claude Code, directly:

claude mcp add trello -e TRELLO_API_KEY=your-api-key -e TRELLO_TOKEN=your-token -- npx -y @delorenj/mcp-server-trello@1.8.1

Cursor, directly, in ~/.cursor/mcp.json or .cursor/mcp.json:

{
  "mcpServers": {
    "trello": {
      "command": "npx",
      "args": ["-y", "@delorenj/mcp-server-trello@1.8.1"],
      "env": {
        "TRELLO_API_KEY": "${env:TRELLO_API_KEY}",
        "TRELLO_TOKEN": "${env:TRELLO_TOKEN}"
      }
    }
  }
}

Run locally, the server also reads TRELLO_BOARD_ID for a default board and TRELLO_ALLOWED_WORKSPACES, a comma-separated list of workspace IDs. That list only filters the tools that list boards and workspaces, pick a workspace or create a board. A tool given a card or list ID from another workspace still works, because the token can reach it.

The client tutorials, How to add MCP servers to Claude Code and How to set up MCP servers in Cursor, explain config scopes and how to keep the token out of files you commit.

Through Statio: install Trello from the catalog, paste the API key and token under the server's Credentials, and grant tools under Permissions. A new server has no grants, so nobody can call it until you add some. The desktop app then writes your client configs. The manual entry is the local proxy URL with the server name from your dashboard, http://127.0.0.1:7777/trello/mcp, added in Claude Code with claude mcp add --transport http trello <url> or as a "url" in Cursor. The hosted server only reaches api.trello.com. Tools missing after that? Start a fresh session, then work through How to debug an MCP server that won't connect.

What to hold back

Start with what the token reaches. It acts as you on every board you're a member of, in every workspace, and the server has no setting that narrows a token to one board. If one agent should only see one board, make a Trello account for it and add that account to that board.

attach_file_to_card and attach_image_to_card accept a file:// path, and the server uploads whatever file it can read to the card. Run locally, that means any file in your home directory, including SSH keys and .env files, ends up on a Trello card where every board member can open it. Remote URLs are checked (HTTPS only, no localhost or private addresses). Local paths are not. Through Statio the server runs in a hosted container with none of your files, but if you run it on your laptop, deny both tools unless the agent's job is attaching files.

archive_card, archive_list and update_list with closed take work off the board. Archived items can be restored, but an agent that archives a list also takes every card in it out of view, and nobody may notice for days. delete_comment, delete_checklist_item and delete_label are permanent. Deleting a label strips it from every card on the board, which breaks any filter or automation built on it.

add_cards_to_list and create_board are how a looping agent makes a mess. One call can create dozens of cards, and nothing stops it running again. A triage agent that needs to file cards can usually do it with add_card_to_list one at a time.

assign_member_to_card, add_comment and move_card act in front of people. Each shows on the card as done by you, and anyone watching the card gets a notification. An agent that assigns ten cards to a colleague does it in your name. copy_card and move_card also work across boards, so they can carry a card's description, attachments and comments from a private board onto a shared one.

Card names, descriptions, comments and checklist items are written by whoever is on the board. Any of that text can carry instructions aimed at the agent. Pairing get_card or get_card_comments with the write tools is what lets that text change your boards.

With Statio you grant each tool separately, so a standup agent can get the read tools and add_comment while a planning agent also gets add_card_to_list and move_card. MCP permissions covers how grants and denies combine.

Questions