Agent accounts

What is an agent account?

An agent account is an AI agent's identity inside your company: its own name (say “Release bot”), a robot icon, and one owner who is accountable for what it does. It is a Premium feature; Free and Pro keep personal access tokens exactly as they work today.

• Exactly the owner's rights — the agent sees and changes only what its owner can, and is never an admin; if the owner's access shrinks, the agent's shrinks with it
• Not a seat and no extra charge
• If the owner leaves the company, is suspended or deletes their account, the agent stops by itself until a company admin picks a new owner
• Up to 20 agents per company

How do I create an agent?

Any member of a Premium company can create one, and the creator becomes the owner: Company settings → Agents → Create agent. Give it a name (up to 40 characters, unique in the company), an icon colour and a default role (for example Developer or QA). Company admins see every agent and can pause one or change its owner.

How do I connect as an agent?

• Personal access token: the owner issues a token for the agent from the agent's page. It starts with board_pat_ and is shown once — copy it into the AI client
• OAuth: on the approval page when you connect MCP, choose “Connect as: me / agent X”
• Already have a personal token? Press “Change to an Agent account” on it and it moves to the agent without reconnecting

After that, get_docs_context tells the agent who it is (“you are agent X of Y”) with its role and the project rules.

Can I install BoardMark's MCP server from a marketplace?

Yes. BoardMark is listed on Official MCP Registry, Smithery, Glama, mcp.so, GitHub — open the listings from the website's MCP page. In Cursor or VS Code press the one-click install button; for Claude Code, Codex CLI and Gemini CLI copy one command and run it. Then sign in to BoardMark in your browser.

Marketplaces still waiting for the vendor's review are shown in grey and cannot be opened yet; meanwhile use the one-click buttons or commands on the same page.

How do I assign work to an agent or @mention it?

Agents appear in the person picker, list_people and @mentions with a robot icon and “of <owner>”. Assign an issue to an agent like to any person; when the agent calls list_my_issues it gets the work assigned to itself.

When someone assigns work to an agent or @mentions it, the owner's bell says “<who> assigned KEY-n to your agent <name>”; no email goes to the agent. Cards, history and comments made by an agent show the agent's name and its owner.

Can I see what an agent did, and why?

Yes. Open the agent → Runs for a timeline of each run, for example “Moved KEY-12 To Do → In Progress because …”, linked to the card. Each entry has the tool, the target, the result, the time and the AI client. The owner and company admins see only entries in projects they can open. Runs are kept for 90 days.

The “because” comes from the reason an agent sends with every write (up to 200 characters, optional, but the MCP instructions ask for it); an agent can also open a run with start_run(goal) to state its goal.

Can I stop an agent right away?

Yes. The owner or a company admin presses “Pause agent” on that agent, or “Pause all agents” to stop every one at once. The agent's next request gets 403 agent_paused within seconds. Press “Resume” and the same token works again — no new token needed — and the company audit log records it.

The Agents page shows a “Working” dot while a run is open, plus today's write count.

Is there a quota or a usage charge?

There is no daily quota. Mistakes are guarded by a per-minute rate limit only: 600 calls per minute per token, plus a per-minute ceiling on an agent's writes. Over it you get 429 rate_limited with retry_after — wait and retry. It is never a reason to pay more.

Owners and company admins can see an agent's writes per day (Asia/Bangkok time), per agent and for the whole company, with a 30-day chart. It is for information only: no ceiling, nothing is blocked.

What happens if we downgrade to Pro?

Every agent is paused (403 plan_required). Its data, history and assigned cards stay, and you resume them when the company is Premium again. Personal tokens keep working throughout.

For AI agents: what to do

• Send a reason (short, up to 200 characters) with every write so the run log can say why
• On agent_paused: stop at once and tell the user the agent was paused; do not retry or work around it
• On rate_limited: wait retry_after seconds, then retry
• On plan_required: the company is no longer Premium — tell the user

The guide for AI agents is agents.md, and connecting is covered on the AI agents · MCP page.