Claude Desktop is the app you already have open next to the board, which makes it a good place to ask what is in flight, write up a ticket properly, or tidy a page of documentation. Agenticboard runs an MCP server it can reach with a scoped token. Claude Desktop does not send headers directly, so the connection is bridged with mcp-remote, a small npm package run through npx.
In Agenticboard, open Settings → API tokens → New token and choose three things: the access level, either Read only for viewing boards, tasks and docs or Read & edit to also create and update tasks, comment, edit docs and attach files; which projects it may touch, all of them or an allow-list; and whether it can read and write documentation. Copy the token when it is shown — it is only displayed once.
Add Agenticboard to claude_desktop_config.json, with your own token in place of the placeholder, then restart Claude Desktop.
{
"mcpServers": {
"agenticboard": {
"command": "npx",
"args": [
"mcp-remote",
"https://agenticboard.app/api/mcp/",
"--header",
"Authorization: Bearer agb_your_token_here"
]
}
}
}Ask Claude to list your projects, or call tools/list. You should see the tools your token is allowed to use: the write tools appear only for a Read & edit token, and the doc tools only when documentation access is on.
If you would rather not keep a token on disk at all, Agenticboard is also a custom connector on claude.ai, which authenticates over OAuth instead — that guide is here. Both paths reach the same MCP server.
The token decides the shape of the toolset. These are the tools the MCP server exposes.
list_projects, get_board, get_task and list_members give it columns, cards, labels, comments, attachments and testing feedback. create_task and update_task open a card or change its title, description, status, priority, type, assignee, deadline, PR URL, version or cycle. get_workflow returns the project's ordered steps, and the write tools return the next step to take, so the work happens in the order you set rather than the order it occurs to anyone. find_resolved searches done tasks and returns how each was fixed, and get_task_history gives one ticket's full trail. list_docs, get_doc, create_doc and update_doc work the same pages your team reads. Edits carry a reason into the doc's version history. comment_task, add_test_report and attach_file put the test result and the screenshot on the card, where a reviewer will look for them. search covers projects, tasks and docs, and get_history with search_history reach the knowledge behind them: doc changes, resolved tickets and captured errors. A token acts as you, narrowed by its scope, and every call it makes is logged under the token's Activity. Revoke it any time from Settings → API tokens; anything using it loses access immediately.
Set up a board, write the first ticket, and ask Claude about it in the app you already have open.
MCP access comes with the Pro plan — see pricing.
Other guides: Claude Code · Cursor · claude.ai · Sentry