For AI agents
An MCP issue tracker your AI agents actually work
Tickhatch is a work tracker with a built-in MCP endpoint. Point Claude Code, Cursor, or any MCP client at it with a scoped tkh_ token and the agent can find, claim, comment on, and move tickets on the same board your team sees. One rule holds: an agent can’t mark work shipped — a merged PR or a human does — so the board stays honest about what’s live.
Connect an agent in one line
Mint a token in Settings → Agent access, then point your client at the endpoint. In Claude Code that’s a single command:
claude mcp add --transport http tickhatch https://tickhatch.com/api/mcp --header "Authorization: Bearer tkh_YOUR_TOKEN"Cursor and other MCP clients take a few lines of config — see the full setup guide.
What an agent can do on your board
Ten tools, small on purpose — enough to run the whole New → In Progress → Testing loop without handing over the keys to everything. A few of them:
next_ticket
The single best unclaimed ticket to work next, priority-sorted, full detail.
claim_ticket
Assign it and move New → In Progress. Fails cleanly if someone got there first.
move_ticket
Advance a ticket — to any status except In Production.
comment
Leave a progress note so the board tells the story.
create_ticket
File a new idea, bug, or task on a board.
board_summary
Cheap orientation: ticket counts per status column.
The guardrail: agents can’t ship
Every token carries a scope — Read, Write, or Write + ship. Write is the default, and the gap between Write and ship is the whole point: an agent can plan, claim, build, and hand work over for testing, but moving a ticket to In Production stays with a merged pull request or a human. That’s what keeps “shipped” meaning shipped — no autopilot quietly marking work done. Pin a token to one board, revoke it on the next request, and read the audit log to see exactly which agent did what.
How it compares
| Feature | Tickhatch | Agent-memory tools (Beads, Taskmaster) | Agent-first boards (Hiveship, TaskPeace) |
|---|---|---|---|
| One board humans and agents share | The agent's private plan | ||
| Hosted — nothing to install | Local files / Git | ||
| Agents can mark work shipped | No — by design | — | Yes (autopilot) |
| Non-technical intake (text → ticket) | — | — | |
| GitHub: merged PR ships the ticket | — | — | |
| Per-board scoped tokens + audit log | — | Varies | |
| Pricing | Free; $10/mo flat workspace | Free / open source | Free + ~$10/mo |
Competitor details from public pages as of mid-2026; products change.
Who it’s for — and who shouldn’t use it
Tickhatch fits a solo dev or a tiny team (one to five) where humans and agents work the same backlog and a non-technical teammate wants to see what shipped. It’s the wrong tool if:
- You want the agent to decompose and remember its own subtasks — use Beads or Taskmaster for that private planning layer.
- You want an agent to deploy end-to-end on autopilot — the ship guardrail is deliberate and won't get out of your way.
- You run cycles, sprints, and roadmaps for a team bigger than five — that's Linear's job, not this.
FAQ
What is an MCP issue tracker?
An issue tracker an AI agent can operate directly over MCP (the Model Context Protocol — the standard way agents connect to tools), instead of you copy-pasting between the agent and your board. Tickhatch exposes a first-party MCP endpoint, so a connected agent can read, create, comment on, and move tickets on the same board your team uses.
Can AI agents create and move my tickets?
Yes. With a scoped tkh_ token, Claude Code, Cursor, or any MCP client can call next_ticket, claim_ticket, create_ticket, comment, and move_ticket — the full New → In Progress → Testing loop. Every call is rate-limited and audit-logged per token.
Can an agent ship to production?
No — that's the deliberate guardrail. A Write token moves tickets between any statuses except In Production. Marking work shipped is reserved for a merged pull request (via GitHub auto-ship) or a human, so the board stays truthful about what's actually live. You can mint a rare Write + ship token if an agent genuinely deploys.
Which agents work with Tickhatch?
Anything that speaks MCP over the streamable HTTP transport with custom headers: Claude Code, Cursor, Codex, and other MCP clients. Setup is one command in Claude Code, a few lines of JSON in Cursor.
How is this different from Beads or Taskmaster?
Those give a single agent structured memory to plan its own long-horizon work — great for that. Tickhatch is a shared team board: the same tickets your cofounders see, with human intake, GitHub auto-ship, and a guardrail that keeps 'shipped' honest. Use Beads or Taskmaster for the agent's private plan; use Tickhatch when humans and agents need one board.
Do humans and agents share one board?
Yes — that's the point. Agents work the exact tickets your team sees on the four-stage board, so nothing lives in a separate agent-only system. A non-technical cofounder can text an idea in and watch it move to shipped.
Where does my data live?
Tickhatch is hosted — your boards live in your Tickhatch workspace, reachable at https://tickhatch.com/api/mcp. Tokens are scoped (read / write / ship), optionally pinned to one board, and stored only as a SHA-256 hash. Revoke any token from Settings → Agent access and it stops working on the next request.
Give your agent a board — and a boundary.
Free for solo devs. Agent access is part of Pro — a flat $10/month per workspace, no per-seat, no per-agent.