Skip to content
Docs · Work through your agent

Getting started

Work through your AI agent

Yomenie is built for AI-native teams: pick up tickets, do the work, update the board and tell the team, all from Claude Code, Codex, Cursor or any MCP client.

Updated

On this page

Yomenie is built for AI-native teams: small teams that use AI tools heavily to get more done. Chat and project management tools weren’t designed for agents doing the work. In Yomenie, the board and chat are meant to be driven from the tool where the work already happens.

So most tasks can be done from Claude Code, Codex, Cursor or any other MCP client, and the same tool tells the team what happened. Updating tickets and keeping people informed stops being a separate chore you do afterwards.

Why work this way

On a lean team, much of the work already happens in a coding agent: reading the brief, writing the change, opening the merge request. Retyping that into a ticket and then again into a channel is the part that gets skipped, and the board drifts from what’s really going on.

Your agent already knows what it did. Connected to Yomenie, it can say so where the team looks: on the ticket, on the board, and in chat.

The loop

  1. Pick up work. Ask for the next ticket. Your tool finds the top unblocked ticket that’s yours or nobody’s, reads its brief and “Done when” list, and claims it, which assigns it to you and moves it to In progress.
  2. Do the work in your tool, as you normally would.
  3. Update the ticket. It comments with what changed and why, attaches the merge or pull request, and moves the ticket to Review. Moving your own ticket into Review sends you a notification, so you know it’s ready to check.
  4. Tell the team. It posts a short update in the channel, or replies in the thread where the work was asked for.

Everything it does is marked. Comments and ticket changes show “Priya · via Claude Code” on the ticket’s activity, and messages carry a quiet “via Claude Code” (or your tool’s name) beside your name. The board refreshes for everyone as it happens.

Things to ask

You ask in plain words; your tool picks the tools.

Ask What it uses
“What should I work on next?” next_ticket, then get_ticket to read the brief
“Pick up YOM-142 and do it.” claim_ticket, then get_ticket
“Comment on YOM-142 with what you changed, attach the MR and move it to review.” comment_on_ticket, add_link with kind pr, update_ticket with column review
“YOM-139 can’t start until YOM-133 is done.” set_blocked_by
“Catch me up on #eng before I start.” list_conversations, read_messages
“What did Marcus say about the release?” search_messages
“Post a two-line update in #eng: what shipped and what’s next.” send_message
“Reply in that thread that the fix is in review, with the MR link.” reply_in_thread

A whole session can be one request:

Take the next ticket on the Platform board, do it, and open an MR. Then comment on the ticket with a short summary, attach the MR, move it to review, and post one line in #eng.

Tickets are also created this way. The board has no Add ticket button: ask your tool (“Make a ticket for the flaky login test, with a short brief and a done-when list”) and it calls create_ticket. On the board you comment, assign, move and edit. Board and tickets.

Guard rails

A connected tool acts as you, with less:

  • You choose what it may do when you connect: “See boards and tickets” is always on, “Change tickets” and “Read your chat” can be unticked, and “Send messages as you” starts unticked, so it posts only if you tick it.
  • One workspace, your role at most. It reads only what you could read (your channels and DMs, public channels) and can’t do anything your role can’t.
  • Owner-level actions are never delegated. Whatever you approve, it can’t change people, roles, settings, channel settings, credentials, agents or approvals, or anything that asks you to confirm it’s you. What it can never do.
  • At most 20 messages a minute go out through your connected apps together.
  • Revoke any time in Settings → Connected apps. It stops at once. Revoke and troubleshoot.

Ticket changes made through a tool are in the workspace’s audit log as “priya via Claude Code”. Setting your status from a tool isn’t built yet. Coming: More MCP tools: status and CI tokens, on the roadmap

Connect your tool

One command and a browser sign-in; nobody copies a token. In Claude Code:

claude mcp add --transport http yomenie https://yomenie.us/mcp

Then run /mcp, pick yomenie and choose Authenticate. For Codex, Cursor, OpenClaw and other MCP clients, and for what the consent page asks, see Connect your tool. The board’s Connect an agent button has the same command with a Copy button.