Taskhold
Server Details
A task list shared by you and your coding agents (Claude Code, Cursor, Codex and any MCP client). Agents plan each task into steps, report live progress to a board, and block on questions you answer from your phone.
- Status
- Healthy
- Uptime
- 100.0% over the last 24 hours
- Last Tested
- Transport
- Streamable HTTP · MCP 2025-11-25
- URL
- ChatGPT
- –
- Claude
- Taskhold
TDQS
Scored across 24 tools
The set covers many distinct facets of task management (task CRUD, steps/planning, blocking, progress, projects, labels, search, updates), and the descriptions actively steer between neighbours (e.g. 'To file and start at once: taskhold_begin', 'use taskhold_create for anything else'). Still, taskhold_begin / taskhold_start / taskhold_next overlap on 'start work', and taskhold_unblock / taskhold_release / taskhold_complete all move status, so an agent must read the prose to pick correctly rather than the names.
Every tool uses the same taskhold_ prefix, snake_case, and a verb-first construction (add_note, create_project, set_priority, get_summary, update_project). Intransitive verbs like begin, block, complete and next still follow the same verb-led convention, so there is no mixing of styles. Names map predictably to resource+action, with only taskhold_updates being a plural noun used as a name.
24 tools is heavy for a task board and pushes past the comfortable range; several are near-duplicates of others (create vs create_from_todo, begin vs start, get vs get_current, list vs list_projects/list_labels). The domain is genuinely rich (steps, blocking, human replies, progress), so most tools earn something, but the surface is larger than it needs to be.
The lifecycle is well covered: create, plan, start, progress, block/unblock, release, complete with structured outcome, notes, search, projects, labels and a change feed. The conspicuous gaps are deletion/archival — there is no tool to delete or archive a task, step, project, note or label — which an agent cannot work around if cleanup is requested.
Available Tools
24 toolstaskhold_add_noteAdd a noteAInspect
Add a note to a task: findings, decisions, what you tried. The person reads these. Notes are append-only and shown oldest first (taskhold_get shows the latest 10). Returns a one-line confirmation; an unknown task_id fails with "Task not found".
Inputs: note (required), task_id (required).
taskhold_beginBegin a taskAInspect
Start work in one call: files a new task (or takes an existing one with task_id), plans its steps, marks it in progress and starts the first step. Use this whenever the person asks for something: it puts the request on their board with a checklist and a progress bar. Returns the task id, the numbered steps with ids, the step now running, project notes and a cursor for taskhold_updates. Every event of the call (created, planned, started) carries the agent label. On a task already under way it appends the steps. Fails without title or task_id, on an unknown task_id (get one from taskhold_list) or on a step (begin its parent, or taskhold_start the step).
Inputs: agent, steps, title, labels, +5 more.
taskhold_blockBlock a task with a questionAInspect
Mark a task (or step) blocked when you can't continue without the person (a decision, missing info, access). Returns the question and the exact taskhold_updates call to wait with. Write the reason as one question they can answer from the board in a line, e.g. "Decide: drop the legacy column or keep it nullable?". When the answer is a choice, pass 2-4 short options (with recommended and if_no_answer): the person answers with one tap and the reply reads Reply to "": chose "". The board shows it under Needs you. Wait for the reply with taskhold_updates {task_id, since, wait: 50}; keep working on anything that doesn't depend on it meanwhile. If the reply doesn't settle it, block again with a narrower question; don't guess. Blocking an already-blocked task asks the new question instead, and the earlier question stays in the task's history; an unknown task_id fails with "Task not found".
Inputs: reason (required), task_id (required), options, recommended, +1 more.
taskhold_completeComplete a taskAIdempotentInspect
Mark a task or step done, with a summary of what you did and, optionally, a structured outcome: files changed (with +/- line counts), checks run (pass/fail/skip), links (PR, commit, branch, deploy, doc), follow-up task ids, risks and how the person can verify it. The person sees it as a What changed card; a step's outcome rolls up to its task. When the person should look at the work, pass needs_review: true with outcome.verify (how you checked it): the task lands in their Ready for review queue. Completing a step tells you the next one; start_next: true also starts it, and finish_task: true on the last open step closes the task too (with the same summary) in one call. For a step it returns steps done, the task's bar and the next step's id (or, after the last step, a reminder to complete the task). Completing a task closes its open steps. Repeating it on a task that is already done changes nothing and keeps the first summary (add more with taskhold_add_note); an unknown task_id fails with "Task not found", an invalid outcome fails without changing anything.
Inputs: task_id (required), outcome, summary, start_next, +2 more.
taskhold_createCreate a taskAInspect
File work for later (a bug, a follow-up, a chore); it lands in To Do. To file and start at once: taskhold_begin. parent_task_id adds one step to a task (taskhold_plan for several). Returns id, status and priority. Fails on an unknown parent_task_id or a step as parent, an unknown project_id (pass project by name) or at the task ceiling.
Inputs: title (required), labels, repeat, project, +4 more.
taskhold_create_from_todoCreate a task from a TODOAInspect
Create a task from a TODO comment in code, keeping the file path and line number (label todo-from-code plus any labels you pass; priority as in taskhold_create; use taskhold_create for anything else). Returns the new task with its id and code location. Fails at the account's task ceiling or on an invalid project name.
Inputs: title (required), file_path (required), line_number (required), labels, +3 more.
taskhold_create_projectCreate a projectAInspect
Create a project (a group of tasks, usually one per repo). Optional: any tool that takes project creates it on first use. A project with this name (any case) comes back unchanged, and the reply says if your description or color wasn't applied. Says whether it was created, restored or already existed, plus its id. Fails on an invalid name or at the project ceiling.
Inputs: name (required), color, description.
taskhold_getGet a taskARead-onlyInspect
Get one task in full: status, priority, project, labels, repeat, time worked and waited on the person, description, numbered steps with ids and summaries, the 10 latest notes, the completion summary and what changed (files, checks, links, risks, how to verify). Unknown ids fail with "Task not found".
Inputs: task_id (required).
taskhold_get_currentGet current tasksARead-onlyInspect
Get the task in progress (in this project when you pass project), with its steps and the next one to do. Call this at the start of a session. Returns the most recently active in-progress task, its numbered steps with ids (with descriptions and the summaries of done steps), the next step's id, project notes and the latest 3 notes (taskhold_get shows 10); with nothing in progress (or an unknown project) it says so and names taskhold_begin and taskhold_next.
Inputs: project.
taskhold_get_summaryGet a summaryARead-onlyInspect
Task counts by status, priority and who added them (human or agent). Returns the total and one list per grouping; counts top-level tasks only, not steps.
Inputs: no named parameters.
taskhold_listList tasksARead-onlyInspect
List top-level tasks, one line each, grouped by status. Pass project to see one repo's work. Each line has status, title, steps done, progress and id; steps aren't listed (taskhold_get shows them). An unknown project name returns no tasks and is not created.
Inputs: limit, status, project, priority, +1 more.
taskhold_list_labelsList labelsARead-onlyInspect
List the account's labels, one name per line. Labels are created by passing them to taskhold_create or taskhold_begin.
Inputs: no named parameters.
taskhold_list_projectsList projectsARead-onlyInspect
List projects with their open work, one line each. You rarely need this: every tool takes project as a name. Each line has the name, open counts (working, blocked, todo) and id. With project: that project's notes.
Inputs: project.
taskhold_nextTake the next taskAInspect
Work the list as a queue: takes the next To Do task (highest priority, then oldest by default) and starts it and its first step. Atomic: no 2 agents get the same task. Returns the task, steps, project notes and next call, or that To Do is empty.
Inputs: agent, project.
taskhold_planPlan task stepsADestructiveInspect
Add ordered steps (subtasks) to a task so the person sees a checklist; the progress bar fills as steps finish. taskhold_begin already does this for new work. For an existing task: taskhold_start it, then taskhold_plan. Steps don't nest. Returns the task's whole step list with ids. Fails on an unknown task_id, on a step (plan its parent) or past 200 steps per task. Planning a done task adds todo steps but leaves it done; taskhold_start it to reopen.
Inputs: steps (required), task_id (required), replace.
taskhold_progressReport progressAIdempotentInspect
Move the person's progress bar during a long step or a task without steps: percent, or current and total (e.g. 12 of 40 files), plus a short message about what's happening now. On a task with steps, the steps drive the bar: numbers you send for the task go to its running step. Call at milestones, every few minutes at most. Returns the task's progress bar and, on a task with steps, which step your numbers went to. Fails without percent, total or message, on an unknown task_id, or past the per-minute limit (the error says when to retry).
Inputs: task_id (required), total, current, message, +1 more.
taskhold_releaseHand back a taskAInspect
Put unfinished work back in To Do, with a note, when you stop and nobody has to answer (e.g. waiting on a captcha). Its steps go too, your label comes off, progress stays. Need the person? taskhold_block. Done tasks are refused.
Inputs: task_id (required), note.
taskhold_searchSearch tasksARead-onlyInspect
Search tasks and steps by text in the title or description (case-insensitive, up to 100 tasks scanned). Returns matching lines grouped by status, with ids (a step names the task it belongs to), or a line saying nothing matched.
Inputs: query (required), status, project, project_id.
taskhold_set_prioritySet priorityAIdempotentInspect
Change a task's priority. Returns the task title and new priority; an unknown task_id fails with "Task not found".
Inputs: task_id (required), priority (required).
taskhold_startStart a taskAIdempotentInspect
Mark a task or step in progress when you begin it. Starting a step also starts its task. For new work, taskhold_begin files, plans and starts in one call. Returns the id, status and the next step's id when it has steps. Safe to repeat; it also resumes a blocked or done task. An unknown task_id fails with "Task not found".
Inputs: task_id (required), agent.
taskhold_unblockUnblock a taskAIdempotentInspect
Move a blocked task back to todo once the blocker is resolved (taskhold_start resumes it directly). Returns the task with status todo. A task that isn't blocked is left as it is and the reply says its status; an unknown task_id fails with "Task not found".
Inputs: task_id (required).
taskhold_updateEdit a taskADestructiveIdempotentInspect
Change a task's title, description, priority, project or repeat; only fields you pass change. Status goes through taskhold_start, taskhold_complete, taskhold_block and taskhold_release. Returns the task. Fails on an unknown task_id or project_id (pass project by name).
Inputs: task_id (required), title, repeat, project, +3 more.
taskhold_update_projectEdit a projectADestructiveIdempotentInspect
Change a project's name, description, color or project notes. Only fields you pass change. Returns the project as it is now. Fails on an unknown project (it is never created here), on a name another project already has, or with nothing to change.
Inputs: project (required), name, color, agent_notes, +1 more.
taskhold_updatesGet the person's updatesARead-onlyInspect
What the person changed since you last looked: replies (notes), new tasks, priority changes, unblocks. Call when you resume, between steps, and after blocking (with wait, until the reply lands). Pass the last cursor you got (from any call) as since. Your own changes are hidden unless include_agent is true. Returns one line per change, oldest first (time, who, task, what, task id), then the next cursor, also when nothing is new. An unknown project returns no changes; an unknown task_id fails with "Task not found".
Inputs: wait, since, project, task_id, +1 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Related MCP Connectors
Task & board management for AI agents + humans. Kanban, comments, digests via MCP.
Project board for ChatGPT, Claude and coding agents, with scoped MCP access to projects and tasks.
AI-native Kanban board — connect Claude to claim, work and move your tasks over MCP.
Shared task board and knowledge base for AI coding agents Give your coding agents a shared task board and knowledge base, so the plan survives between sessions and across agents.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceA shared task board exposed over MCP that lets a team's mixed AI coding agents (Claude Code, Codex, Copilot, Gemini CLI, Cursor) claim tasks, build each one on its own git branch, review and merge pull requests, and coordinate through board activity and locks. Humans watch and steer the same board from a web page, answering questions, approving tasks and choosing who merges.MIT
- AlicenseNot gradedqualityBmaintenanceDesktop task board that coding agents keep themselves: they register goals, plan steps and progress over MCP, and wait for your sign-off. Local SQLite, three tools.2MIT
- AlicenseAqualityBmaintenanceLocal-first task and progress database shared by developers and coding agents. Agents claim ready tasks over MCP, record progress and handoffs,and submit commits for independent review; developers confirm, prioritize, and approve work in a local web dashboard. Data stays in a SQLite file inside your project, with no cloud service or account required.15269 npm13Apache 2.0
- AlicenseNot gradedqualityDmaintenanceMCP server for task management that enables AI agents to read, create, update tasks, and track work sessions, allowing agents and humans to collaborate on the same task board.14MIT
Glama MCP Gateway
Add one secure layer between your agents and this server.