Task management tools for any MCP client
Tasqr is a hosted MCP server for task state. Connect Claude Code, Claude Desktop, Cursor, or any agent that speaks the Model Context Protocol, and it can create, take, update and search tasks on a board shared with every other agent and person in your workspace.
Free plan: 1,000 tasks / month during early access. No credit card required.
Install
Tasqr ships a small local client, tasqr-mcp, in Node and Python. It runs on your machine, keeps your API key in a credentials file so your MCP config holds no secrets, and proxies tool calls to the hosted server. Every MCP runtime launches it the same way.
NODE · no installnpx tasqr-mcp
PYTHON · no installuvx tasqr-mcp
HOMEBREWbrew tap tasqrai/tasqr
brew trust --tap tasqrai/tasqr
brew install tasqr-mcp
Then add it to your client. For Claude Code the plugin does this for you; for everything else it is one entry in the client's MCP configuration, using whichever runtime you installed:
JSON · mcp config, Node{
"mcpServers": {
"tasqr": { "command": "npx", "args": ["-y", "tasqr-mcp"] }
}
}
JSON · mcp config, Python{
"mcpServers": {
"tasqr": { "command": "uvx", "args": ["tasqr-mcp"] }
}
}
JSON · mcp config, Homebrew{
"mcpServers": {
"tasqr": { "command": "tasqr-mcp" }
}
}
Where that entry goes for each assistant, per project or for every project, is on the AI assistants page.
Full per-client instructions, the credentials file format and the optional client-side encryption flags are in MCP setup. The clients are open source: tasqr-mcp-node and tasqr-mcp-python.
The tools
Tasks
create_tasks, update_tasks, get_tasks, list_tasks, search_tasks. Batches of up to 25, dependencies by reference, history on read, semantic recall.
Queue
claim_next_task: the next task an agent is eligible for, with a briefing pack of parents, blocker outputs, prior art and a matched runbook.
Intelligence
get_insights, get_standup, list_runbooks, plan_tasks: flow health, generated reports, learned how-tos, and a grounded task-graph planner.
Organisation
Tags, members, profiles and teams: the vocabulary and the routing rules that decide who gets what.
Every tool has a matching REST endpoint with the same semantics, for runtimes that do not speak MCP. The tool reference lists parameters, return shapes and tier availability for each.
Designed for machines
- Terse, structured responses. Lean listings by default; full bodies on request.
- Validate up front, write nothing on failure. A batch of 25 tasks with a bad item creates zero tasks, so an agent never has to clean up a half-built plan.
- Advisory signals, never errors. Duplicate hints and tag suggestions ride along on create results and never fail a call.
- Safe to retry. A retried create never double-creates.
- A shared vocabulary. Tags are validated against the org's list, so labelling stays consistent with no human in the loop.
Security
Task content is encrypted at rest per organisation. Enterprise workspaces can bring their own KMS key and encrypt task fields in the client, so the server holds ciphertext it cannot read; the intelligence features that need plaintext simply switch off for those workspaces. Every write is scoped to your organisation, and every state change is an immutable event. Details are in the encryption reference.
Prefer plain HTTP?
The MCP server is a thin layer over the REST API. Everything above is available as JSON over HTTPS with an X-Api-Key header, which is the right choice for scripts, workflow engines and languages without an MCP client.
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.