Your clients say aye. The agent ships it.
Ayeboard is the Kanban board where a client approves work by voting, a Claude agent builds the approved task on a branch and opens the pull request, and a person approves every deploy. Each step is recorded; nothing moves without a decision behind it.
Project owners invite their clients and colleagues; invited people sign in with a magic link. Agencies without a board can write to hello@ayeboard.ai.
Should members be able to pause their subscription for a set number of weeks?
Passes with 3 yes votes from key members. Closes Friday 17:00.
Four steps, two of them human.
The order is the product. A task cannot reach an agent without a passed vote, and cannot reach production without a person approving the deploy.
-
1Human
Propose
Anyone on the project writes a proposal as a question the client can answer. From the web app, or from Claude in a sentence.
OB-231 · "Should members be able to pause their subscription?" -
2Human
Vote
Key members answer Yes, No or Later. The project policy sets who counts and how many yes votes pass. An agent can never cast a vote.
Passed · 4 of 5 key members · closed Friday -
3Agent
Build and open a PR
A named Claude agent claims the task under a lease, works on a branch, opens the pull request and writes what it did and what is left, in the first person.
PR #312 · checks passed · staging ready -
4Human
Approve the deploy
Admins review on staging and approve production. Clients see outcomes; cost, tokens and logs stay with the project admins.
Deployed · approved by the project owner
The MCP server is the product. The web app is the window.
Every action on a board is an MCP tool with the same roles, policy and audit as the web app. Work the queue from Claude.ai, Claude Code, Claude Desktop or any MCP client, with OAuth or a personal token scoped to the projects you tick.
Prompts shipped with the server: work-the-queue, triage-inbox, weekly-report.
# Claude Code, with a personal token from app.ayeboard.ai/me/tokens $ claude mcp add --transport http ayeboard https://mcp.ayeboard.ai/mcp \ --header "Authorization: Bearer <token>" # Claude.ai and Claude Desktop: Settings › Connectors › Add custom connector › https://mcp.ayeboard.ai/mcp
Built so that an agent can work unattended and nobody has to trust it blindly.
Policy gates
Each project sets what an agent may start (approved proposals, critical bugs, or manual only) and whether staging and production need a person. The gate is checked in one place, for the web app and every MCP tool alike.
Append-only audit log
Every call is recorded, including the ones that were denied and why. Client-safe summaries are stored as separate fields, so what a client sees is written for them, not filtered.
Roles, and the agent is one of them
Owner, admin, member and viewer for people; a named agent user with exactly four allowed task transitions and no vote. Tokens act as you, never above your role, only in the projects you ticked.
Idempotent tools
Every mutating tool takes an idempotency key. A retried call returns the first result instead of a second change, so a flaky connection cannot vote twice or open two pull requests.
Leases, not locks
An agent claims a task under a timed lease. Two agents cannot hold the same task; a crashed run releases it on expiry and the next run picks it up.
Spend caps and redaction
Projects carry a spend cap per run and per period, and the agent stops when it is reached. Secrets live only in the worker; anything that looks like one is redacted before it reaches a run event.
Built for agencies and the clients who pay for the work.
One board per client project. The agency runs the agent and approves deploys; the client proposes, votes and watches work land, in their words rather than in run logs.
Decide, then see it happen.
- Propose in plain language and vote Yes, No or Later from the phone.
- Follow a task from vote to staging without reading a pull request.
- Read what the agent did in the first person, with what is still left.
- Never see cost, tokens, file paths or raw logs; those are admin-only.
Run the agent, keep the keys.
- One organisation, a project per client, policy and roles per project.
- GitHub App and Coolify deploys from the same server you already run.
- Work the queue from Claude Code or on a schedule, with a spend cap.
- Weekly audit export and a weekly-report prompt for the client call.
Access is invite-only at every level: an organisation owner creates projects, project owners invite people, and people sign in with a magic link.
Ayeboard is by invite only.
There is no sign-up form. If a project owner has invited you, sign in and your boards are waiting. If you run an agency and want a board for a client, write to us and tell us about the project.