Docket

The issue tracker your AI agents can actually use.

Issues, boards and docs, with a clean web UI for you and an MCP server for Claude and other agents. Same tracker, both at the same time, live.

One container. One SQLite file. Your people and your agents, each with their own sign-in. No SaaS.

Why Docket

Agents are good at doing work and bad at keeping track of it. Docket gives them a place to do that: they pick up issues, post progress, write the spec and move things to review, while you watch it happen in the browser.

• Built for agents and humans together. 22 MCP tools for issues, comments and docs. Every agent gets its own token and name, and claims issues as a delegate, the way Linear's agents do. What an agent does shows up in your UI right away over WebSocket.

• Linear's model, tiny code. Workspaces with members and admins, teams, list and board views, priorities, labels, sub-issues, blockers, keyboard shortcuts (C, /, ⌘↵).

• Docs next to your issues. Markdown docs with version history. Write API-1 and it links to the issue, with its status shown inline.

• Yours. Self-hosted, a single SQLite file, five runtime dependencies. Back it up live with ./backup.sh.

• Works everywhere. Install it as a PWA on iPhone, iPad or Mac. It works offline for the issues and docs you've already opened.

Related MCP server: Yandex Tracker MCP

Quick start

git clone https://github.com/youssefaltai/docket && cd docket
mkdir -p data && sudo chown -R 1000:1000 data
docker compose up -d --build

Open http://localhost:7100. Your data lives in ./data, a single SQLite file. Don't cp it while Docket is running — WAL mode makes that unsafe. Use ./backup.sh instead: it takes a consistent snapshot with VACUUM INTO, safe to run live.

The chown matches ./data to the container's non-root user (bun, uid 1000), which owns it inside the image. Already running Docket without it? Same command, run once, fixes an existing deployment too.

The first time it starts, Docket prints a one-time setup code (docker compose logs docket). Enter it at http://localhost:7100/setup to create your account; you become the admin of your first workspace.

Then give Claude Code its own token: Settings → Workspace → Add agent shows a ready-to-paste command:

claude mcp add --transport http --scope user docket http://localhost:7100/mcp \
  --header "Authorization: Bearer dk_…"

Try: "Create a team called Website in Docket and file issues for everything in TODO.md."

Going further

The container listens on 127.0.0.1:7100 only. Put it behind whatever you already use for HTTPS: a reverse proxy (Caddy, nginx, Traefik), a tunnel, or a private network like Tailscale or WireGuard.

Back up with ./backup.sh. Run it from a nightly cron: it writes a consistent snapshot into data/backups/ and keeps 14 days.

The assistant is a separate service, docket-chat, next to Docket on the same host. Docket's compose creates a Docker network, docket, and docket-chat joins it. Docket reaches it at http://docket-chat:7110, and it reaches Docket at http://docket:7100. Both stay published on 127.0.0.1 only.

cd /opt/apps/docket
echo 'CHAT_URL=http://docket-chat:7110' >> .env
docker compose up -d                       # Docket first: it creates the network
cd /opt/apps/docket-chat                   # then docket-chat (see its README)
mkdir -p data && sudo chown -R 1000:1000 data
docker compose up -d --build

Back up its conversations nightly with Docket's script:

19 3 * * * /opt/apps/docket/backup.sh /opt/apps/docket-chat docket-chat /data/chat.db >> $HOME/docket-backup.log 2>&1

./backup.sh
git pull
docker compose up -d --build

Schema changes apply by themselves on startup and never drop data. Keep the backup until you know the new version works: it is your way back.

Docket copies Linear's model: everyone signs in, each workspace has its own members, and agents are apps with their own tokens.

• People join with an invite link (Settings → Workspace → Invite): whoever opens it creates an account, or joins with the one they're signed in to. There are no passwords and no email: to sign in on a new device, open Settings → Account → Sign in on another device on one where you're signed in (no one else can sign you in, not even an admin). Links work once and expire after 15 minutes.

• Agents are added by an admin (Settings → Workspace → Add agent) and get a token, shown once. They write under their own name, and claiming an issue makes them its delegate while a person stays the assignee.

• Scripts use personal API keys (Settings → Account → API keys), read-only or read-write. Keys can't create other keys, invites or sign-in links: that takes the web app.

• Removing someone is suspending them: their access to the workspace ends at once and their history keeps their name. If it was their only workspace, their sessions and API keys are deleted too, and reinstating them means they sign in again.

Serve it over HTTPS anywhere but localhost. Give each Docket its own hostname: browsers share cookies across ports, so two Dockets on one host (say localhost:7100 and localhost:7200) sign each other out, and any other app on that host can read the session cookie.

Locked out? On the server, docker compose exec docket bun run sign-in-link <username> prints a one-time sign-in link. Set DOCKET_URL so it points at your public address.

In Docker, set these in a .env file next to docker-compose.yml (see .env.example). PORT there only changes the host-side port mapping; the container always listens on 7100 internally.

You can also use an optional config file at $XDG_CONFIG_HOME/docket/config or $XDG_CONFIG_DIRS/docket/config, with KEY=VALUE lines (# starts a comment line; an unquoted value drops a trailing # comment; surrounding quotes are stripped). Real env vars win over the file, unless a var is set but empty — e.g. docker-compose's ${DOCKET_HOSTS:-} — which counts as unset.

list_workspaces, create_workspace, update_workspace, list_members, list_teams, create_team, update_team, list_issues, list_labels, get_issue, create_issue, update_issue, claim_issue, comment_issue, list_documents, get_document, create_document, update_document, comment_document, delete_document, update_comment, delete_comment.

The full REST API and data model are in SPEC.md.

Contributing

Contributions are welcome, from typo fixes to new features. Docket is small on purpose, so you can read the whole codebase in an afternoon.

bun install
bun run dev   # http://localhost:7100, hot reload, data in ./dev.db, setup code DEVEL-SETUP
bun test

See CONTRIBUTING.md for how the code is laid out and what makes a PR easy to merge. Not sure where to start? Open an issue and say hi.

License

MIT