Tasks that outlive the session
Claude Code, Cursor, Claude Desktop, Amazon Quick, Google Antigravity, Amazon Kiro: every assistant forgets the plan when the conversation ends. Tasqr gives them a task list that persists across sessions, machines and assistants, reached over MCP from whichever one you use, so tomorrow's session picks up where today's stopped.
Free plan: 1,000 tasks / month during early access. No credit card required.
Why an assistant needs a task board
Most assistants keep an in-session todo list. It is a good scratchpad, and it stops at the edge of the session: it knows nothing about the subagents, the second laptop, the other assistant you switch to, or the teammate who picks the work up tomorrow. The moment work is measured in sessions rather than minutes, the plan has to live somewhere the next session can read it.
Tasqr is that somewhere. A task carries its title, description, priority, tags, dependencies, whoever holds it, and whatever the last agent produced. Tomorrow's session, a different model, or a different assistant reads the same record and continues. Nobody re-explains the project to start it again.
It is not only for code. A product manager in Claude Desktop or Amazon Quick tracks the launch checklist, the vendor follow-ups and the customer interviews on the same board the engineering agents work from, and the standups cover all of it.
Connect your assistant
Tasqr is an MCP server, so any MCP-capable assistant connects the same way: install the small local client, add one mcpServers entry, and the tasqr tools appear. The entry is identical everywhere; only where it lives differs. Most assistants accept it at two levels, per project or for every project.
JSON · mcpServers entry{
"mcpServers": {
"tasqr": { "command": "uvx", "args": ["tasqr-mcp"] }
}
}
| Assistant | Per project | Every project |
|---|---|---|
| Claude Code | .mcp.json in the project, or install the plugin | ~/.claude.json via claude mcp add --scope user |
| Cursor | .cursor/mcp.json | ~/.cursor/mcp.json |
| Google Antigravity | .agents/mcp_config.json | ~/.gemini/config/mcp_config.json |
| Amazon Kiro | .kiro/settings/mcp.json | ~/.kiro/settings/mcp.json |
| Claude Desktop | Global only | claude_desktop_config.json (Settings › Developer › Edit Config) |
| Amazon Quick Desktop | Global only | Settings › Capabilities › MCP › Add MCP, as a local command or by importing a Kiro or Claude Code config file |
The first time the client starts without a key it walks you through signup and stores the key in a credentials file, so the config holds no secrets. Full details, including Homebrew and Node installs, are in MCP setup.
Claude Code: the plugin
For Claude Code there is a plugin that bundles the MCP client with a skill and installs in two commands:
CLAUDE CODE · install/plugin marketplace add tasqrai/tasqr-claude-code-plugin
/plugin install tasqr@tasqr
The status line is a Claude Code extra too: it shows the task you are on, what the queue would hand you next, and what is blocked, and it is read-only. Setup is in the docs.
Teaching the assistant when to use it
An MCP server gives an assistant tools; it still needs judgement about when to reach for them. The plugin's skill supplies that for Claude Code, and it is plain markdown you can adapt into any other assistant's rules or instructions. What it teaches:
- Plan substantial work as tasks with dependencies. A migration, a feature build, a launch or an audit becomes a set of tasks wired together in one
create_taskscall, so the plan exists before the first step. - Resume instead of duplicating. At session start it looks for tasks already assigned to it, in progress or pending, and continues them.
- Take the next task, with its context.
claim_next_taskhands it the next eligible task along with the parent chain and the outputs of the work it depended on. - Model subagent fan-out as child tasks, so a tree of parallel work is visible on one board and each subagent closes its own task.
- Leave the in-session todo list alone for small things. Quick lookups and short in-session work never become Tasqr tasks. A board cluttered with trivia is worse than no board.
- Track non-code work you ask it to. "Follow up with the vendor Friday" is a first-class task.
What Tasqr adds that a file cannot
Duplicate detection
Creating a task returns open tasks that already match it, so two sessions never grind on the same thing.
Semantic search
search_tasks recalls past work by meaning, including completed tasks' output, so the assistant reuses prior work.
A briefing with every task
Every task an assistant takes arrives with the parent chain, the outputs of its blockers, and prior art. No follow-up fetches to rebuild context.
Standups written for you
Daily, weekly and monthly reports of what your agents did, generated from the task record.
One board, several people
A Tasqr workspace is shared. Your sessions, a colleague's in a different assistant, and any other agent or script that holds a key all read and write the same tasks, with roles, a tag vocabulary and an immutable history of who changed what. See multi-agent coordination for how a fleet of agents shares one queue without colliding.
Connect an agent in a few minutes
Sign in with GitHub, copy your API key, and add the Tasqr MCP server to Claude Code, Claude Desktop, Cursor or any MCP client. The free plan gives you 1,000 tasks a month during early access. No credit card required.