AI access (MCP)

What it is

G Kanban runs a read-only MCP (Model Context Protocol) endpoint. Your own AI agent connects and reads your synced boards. It can never change them. No AI runs on G Kanban’s side; board contents never pass through a model we operate.

What you need

Both of these:

• A Pro subscription, and
• Cloud sync turned on.

MCP reads your synced boards. A free or local-only account has nothing in the cloud to read, and the endpoint returns a clear “requires Pro” error rather than an empty list.

Create a token

Tokens are managed in G Kanban under Settings:

1. Go to Settings → AI access (MCP).
2. Give the token a name so you can tell it apart later (for example, “Claude Desktop” or “Claude Code”).
3. Click Create token.
4. Copy the token immediately. It starts with gk_ro_ and grants read access to all your synced boards. Treat it like a password.

Your token is shown once

Your token is only shown once when generated. Copy it somewhere safe before closing the dialog. If you lose it, you can revoke it anytime and create a new one.

The dialog also gives you a ready-made configuration snippet and a claude mcp add command with your actual endpoint URL already filled in.

Connect your client

The examples below use placeholders (<project-ref>, gk_ro_<your-token>). The snippet and command you copy from the app have the real endpoint URL and your token already filled in, so copying from there is the easiest route.

Claude Code

Run the claude mcp add command in your terminal:

claude mcp add g-kanban "https://<project-ref>.supabase.co/functions/v1/mcp" \
  --transport http --header "Authorization: Bearer gk_ro_<your-token>"

Then in a Claude Code session, run /mcp to confirm g-kanban is listed as connected.

Claude Desktop / Cursor

Claude Desktop and Cursor connect using npx mcp-remote (which requires Node installed). Paste the JSON snippet below into your client’s MCP configuration file (for Claude Desktop, this is claude_desktop_config.json):

{
  "mcpServers": {
    "g-kanban": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://<project-ref>.supabase.co/functions/v1/mcp",
        "--header",
        "Authorization: Bearer gk_ro_<your-token>"
      ]
    }
  }
}

Restart your client after saving the configuration.

What you can ask

Once connected, your AI agent can query your boards using five read-only tools (list_boards, get_board, get_card, search_cards, and board_stats). Here are examples of what you can ask:

Prompt	Tool	What it returns
”What boards do I have?”	list_boards	Your board titles and card/column counts
”What’s in my Backlog, and flag the critical-priority cards”	get_board	Column cards with critical priority cards singled out
”What’s stale on <board>?”	board_stats + get_board	Aging-WIP count, oldest card age, and per-card freshness (fresh, aging, stale)
“How many cards did I finish in the last week?”	board_stats	7-day throughput (throughput7d)
“Find every card that mentions ‘invoice‘“	search_cards	Title and description substring matches across boards
”Give me the full description and checklist of ‘<card title>‘“	get_card	Full description, checklist items, and column history
”Which cards are overdue?”	get_board	Cards evaluated by the agent using the dueDate field

Errors you might see

If something goes wrong while connecting or querying the MCP endpoint, here is what the error responses mean:

Situation	Response
Valid Pro token	200, JSON-RPC result with your data
Missing / wrong / revoked token	HTTP 401, error.code -32002 (all three look identical — by design)
Non-Pro account	HTTP 403, error.code -32003 (“requires Pro with cloud sync”)
Unknown tool name	JSON-RPC error -32602
Chatty client hitting the throttle	HTTP 429, error.code -32029 (generous per-minute limit)

Revoking and rotating tokens

• Revoking a token: In G Kanban, open Settings → AI access (MCP) and click Revoke next to any token. Revocation takes effect immediately; the next request made with that token returns HTTP 401.
• Rotating a token: Create a new token in Settings, update your client’s configuration with the new token, and then revoke the old token.
• Tracking usage: The Last used timestamp on each token row in Settings helps you spot tokens you no longer need.

Limits

• Read-only (no writes yet): The MCP endpoint is strictly read-only. Your agent can read boards, columns, and cards, but it cannot create, move, edit, or delete cards or modify your board settings.
• Card descriptions: get_board omits card descriptions on purpose to keep response payloads compact and reduce token cost for your AI client. To retrieve a card’s full description, checklist, and history, your agent calls get_card.