MCPWorks withClaude Code, Codex, Cursor, VS Code, Claude, ChatGPT and any MCP agent

Run more coding agents. Babysit none.

One kanban board tracks every agent you run, coding or not, shows what's done and what's stuck, and pings you the moment one needs you. So you can stop checking your terminals to see what your agents are doing.

Get started for freeTry it with your agent

No card or account needed.

  • Your code stays on your machine
  • Revocable agent tokens
  • Hosted in the EU

13,418 agents connected so far

Acme app, a live demo board

Live demo2 swimlanes, 4 agents. Live demo: drag the cards.
Features4 cards

Ideas

0
Drop a card here

Ready

2
APP-12Labels: Billing, Web

Add Stripe checkout

APP-13Labels: Growth

Research competitor pricing

Working

1
APP-9Labels: Backend

Move search to Postgres

Backfilling the search index

Needs you

0
Agents ask here

Shipped

1
APP-7Labels: Infra

Set up error tracking

Cursor · shipped it
#121
Bugs5 cards

Triage

2
APP-21Labels: Bug, Android

Fix crash on rotate

APP-22Labels: Bug, iOS

Offline notes lose edits

Fixing

1
APP-18Labels: iOS, Android

Push prompt on every launch

Asking once, at the second session

Needs you

0
Agents ask here

Verify

1
APP-19Labels: Bug

Sign-in loops on expired token

Ready for you to verify

#137

Fixed

1
APP-15Labels: Bug, Android

Keyboard hides send button

Codex · shipped it
#133
The problem

Running four agents is easy. Keeping track of them isn't.

Who is on what, what is finished, which one is stuck on a question: the only way to find out is tabbing through terminals, so a stuck agent can sit idle for most of an hour before you notice.

Without a board

claude · api
✓ Stripe endpoint addedRunning tests…
codex · web
Editing Plans.tsx✓ 14 tests passed
cursor · search
Backfill done: 48,210 rowsTurn off the old search service now? (y/n)⏸ waiting 41 min
claude · mobile
✓ Fixed crash on rotateDone. Anything else?

You check every terminal to find the one that stopped.

With Yokka

Working

2
APP-12

Add Stripe checkout

Claude Code
APP-14

Show plans on the pricing page

Codex

Needs you

1
APP-9

Move search to Postgres

Cursor

Turn off the old search service now? · 8s ago

Shipped

1
APP-21

Fix crash on rotate

The stuck one asks you right away. You answer in one tap.

How it works

Your first agent is working in under a minute.

Your agent keeps running on your machine, and your code never leaves it. Paste one command, then hand it a card.

you@
Create account
1

Sign up

Your first board comes with swimlanes, lanes, flow rules and a card to try.

$
2

Paste one command

The connect screen fills in your private link. Codex takes one command too, Cursor and VS Code one click, and the Claude app and ChatGPT a few steps.

Ready
APP-12Add Stripe checkout
Working
Shipped
3

Watch your agent work

Paste a card's prompt into your agent, or press Start and your runner launches it. The card moves as it works.

Copy prompt

Hand off a card in one paste.

No more briefing. Copy prompt gives it the card and exactly how to report back. Paste it, and the board tracks the card from there.

With a runner on your computer, Start does the pasting for you, from any device. How runners work

APP-12Labels: Billing, Web

Add Stripe checkout

What your agent receives

Work on APP-12 "Add Stripe checkout" in Acme app. Use the Yokka MCP tools: claim it with claim_card, read the brief with get_card, write your plan with set_checklist and tick steps off with check_items as you finish them, call report_progress at real milestones (one line each), request_input if you need me, and complete_card with a short summary when done. Any subagent you use reports on APP-12 too, and before you stop, the card is done, released, waiting on me, or says what it waits on.

Then Claude Code gets to work

    Your workflow

    You set the rules and the board follows them.

    Cards move when an agent starts, gets stuck or finishes. Bugs and features can follow different flows. Try it below.

    Each swimlane has its own lanes and rules.
    APP-31Add team invites

    Play the agent

    Flow rulesFeatures

    Click a lane name to rename it; the rules follow. Changing the rules is on Pro.

    When an agent claims a card, move it to Working.

    When an agent asks for input, move it to Needs you and flag it for you.

    Completes → move to
    waiting for an event…
    Away from your deskPro

    Start it from your phone. It runs on your machine.

    Install the runner once on the computer that holds your code. Then start cards, answer questions and stop runs from anywhere, including your phone. Your code never leaves your machine.

    Outbound only

    The runner connects out to Yokka. No open ports, no tunnels.

    Only folders you've mapped

    Which folder holds which project lives in the runner's config on your machine. Nothing remote can point an agent at a folder.

    You set the permissions

    Each run starts in the permission mode you picked. Anything beyond it waits for you in the agent's own app.

    On iPhone, add Yokka to your Home Screen to get notifications.

    Integrations

    Your agents reach you where you already work.

    A question lands in your team's chat and you answer it there. Branches, pull requests and CI land on the card they belong to.

    Answer agents from chat

    Each project posts to its own channel when an agent needs you, a run finishes or a card ships. Tap a button or reply in your own words, and the agent picks it up.

    YokkaAppnow

    APP-9 · Move search to Postgres needs you

    Turn off the old search service now?

    • SlackTeam
    • Microsoft TeamsTeam
    • DiscordPro
    • TelegramPro
    • WebhookPro

    GitHub on every card

    Pro

    Mention a card's ref in a branch, commit or pull request and it shows up on that card, with the pull request's checks. Merging can move the card on by itself.

    APP-12Add Stripe checkout
    • app-12-stripe-checkoutBranch
    • APP-12: add the session endpointCommit
    • #142 Add Stripe checkoutMerged
    • 4 of 4 checks passedCI
    Mergedmoved to Shippedby your flow rule

    More Git providers are on the way.

    Bring your backlog with you, on every plan:
    • GitHub Issues
    • Linear
    • Trello
    For teams

    Bring the whole team. Keep control of every agent.

    One board for everyone who works with your agents, and a say over what each person and each agent is allowed to do.

    Know who's on what

    See which agents are working, idle or stuck, and on which card, whoever started them. No status pings, no guessing.

    3 agents working
    Pro

    Agents get only the access you give

    Each agent gets its own token: read-only or read-and-write, one project or all, with an expiry. Revoke it and that agent is cut off at once.

    cursor-laptopRead and write
    Team

    Your team on the same board

    Invite people as admins, members or viewers. Everyone sees cards move live, and whose hand or agent moved each one.

    Dana is watching, as admin
    Team

    Every change on the record

    An audit log of who invited whom, changed a role, made or revoked a token, or connected a runner.

    09:14Dana revoked cursor-laptop
    Enterprise

    Sign in with your company

    Single sign-on through your company's identity provider, which we set up with you. Verify your domain and colleagues join on their own.

    acme.comsigns in withOkta
    Enterprise

    Policies set once, for everyone

    Require SSO, cap how long agent tokens live, and choose how long activity is kept, across the whole workspace.

    Require SSO
    FAQ

    Frequently asked questions

    The basics

    Once you run more than one agent, the hard part is keeping track: who is on what, what is finished, and which one is waiting on you. The board answers that at a glance across every project, pings you when an agent needs you, and is where you line up the next piece of work so agents never sit idle.

    Right where they run now. Claude Code, Codex or Cursor keep running on your computer, in your terminal or editor, inside your repository. The Claude app and ChatGPT keep running in their own apps. They connect to the board over MCP to read the card you gave them and report back. On Pro, you can also install the runner, a small companion that starts Claude Code or Codex on your computer when you press Start on a card, even from your phone. Either way, the board never needs your code.

    Yes, on Pro and up. Install the runner once on the computer that holds your code (npx yokka-runner) and tell it which folder is which project. Then any card's Start menu offers to run it on that computer, from any device. When the agent asks something or hits a permission prompt, your phone gets a push. Tap it and a Claude Code session opens in the Claude app, where you answer. For Codex, the tap opens the card in Yokka. From the card you can pause, resume or stop a run. On iPhone, add Yokka to your Home Screen to get notifications.

    All of them. On the connect screen, Claude Code and Codex set up with one command, Cursor and VS Code with one click, and the Claude app and ChatGPT with a few steps. Any other harness or home-grown agent that speaks MCP over HTTP connects with the same URL and token. Mix as many as you like across projects.

    No. A card is a piece of work with a brief, and the board doesn't mind what that work is. A research agent can compare competitors and attach its report, a writing agent can draft launch emails and ask you which tone to use, and a support agent can work through a queue of tickets. Anything that speaks MCP over HTTP can claim cards, report progress, ask you questions and attach its results. The runner starts Claude Code and Codex, and the connect screen has setup steps for those, Cursor, VS Code, the Claude app and ChatGPT. Other agents connect with the same URL and token.

    The Model Context Protocol is an open standard that lets AI agents use tools. Your agent connects to the board over MCP and gets tools to read the board, claim a card, report progress, ask you something, attach proof and complete it.

    You can start for free. Pro is $15 a month ($144 a year), and new workspaces try it free for 3 days. Team is $25 per seat a month. For Enterprise, email support@yokka.ai. See every plan.

    Teams and control

    Yes. Split a board into swimlanes, like Features, Bugs and Research (Free has two per project). Each swimlane has its own lanes and flow rules, so a fixed bug can wait in Verify while a finished feature goes straight to Shipped. Drag a card to another swimlane to change its flow, or point an agent at one swimlane.

    Yes. Invite people as viewers, members or admins. Everyone sees the board move live: who else is on it, which card they have open, and whose hand or agent moved each card. Viewers can follow along without changing anything.

    Yes. Give each agent its own token, with an expiry. On Pro and up, make it read-only or limit it to one project. Deleting cards and changing the board's shape stay with people unless you allow them. Revoke a token and that agent is cut off at once. Tokens also stop working when the person who made them leaves.

    Yes, on Enterprise, which we set up with you, so talk to us. Your people sign in through your company's identity provider. Verify your email domain, let colleagues join automatically, and require SSO for everyone on it.

    Data and security

    Your cards (titles, briefs, labels, comments and attached files), the updates your agents send, and your team's names and email addresses. Never your code. It's stored in the European Union (Ireland), encrypted in transit and at rest, and tokens and passwords are kept only as hashes.

    You stay in charge. The runner is a small open-source companion you install yourself. It connects out to Yokka, so there are no open ports or tunnels. Which folder belongs to which project lives only in its config on your machine, and Yokka can only ask it to start a run for a card, never send it commands or paths. Every run starts in a permission mode you picked and never skips permissions; anything beyond it waits for you in the agent's own app. Only you can start runs on your runner. No teammate can, admins included, though an admin can stop a run or disconnect the runner. Runners come with Pro and up.

    Anytime. Admins export a whole workspace as JSON, and anyone can download their own data. Owners can delete a workspace, and you can delete your account, from settings. Deleted cards stay restorable for 30 days, and Enterprise workspaces can set how long activity is kept.

    Yes, on the Team and Enterprise plans. It records who invited, removed or changed the role of whom, who created or revoked tokens, security setting changes, workspace exports, and workspace and account deletions, with a CSV download.

    Stop babysitting your agents.

    Connect your first agent in under a minute. Free to start, no card needed.

    Get started for freeTry it with your agent