Put AI agents to work on your boards

Adds an item to a task's checklist: at the end, or with position "next" where the work stands (right after the last done item, before the next open one) — use "next" for a step that comes up while working, such as a Fix: item. Items added once the work has started are marked addedLive. Needs tasks:write.

Posts a Markdown comment on a task, signed by this bot. Needs comments:write.

Records a choice the task needs: a question, its options and the recommended one. Mark it critical when a wrong pick would be costly: an unanswered critical decision holds the task back (a running Taskgert does not hand it out, and the people are notified). Otherwise, whoever works the task takes the recommended option if nobody answers. Needs tasks:write.

Appends tasks to the end of a Taskgert's plan. A task can belong to one Taskgert at a time.

Appends a template's checks to a task in one call, in order, leaving out those it already has. Use the project's templates (get_checklist_vocabulary) rather than adding standard steps one by one. Needs tasks:write.

Adds and/or removes assignees of a task. Assignees must be members of the project; "me" is this bot. Needs tasks:write.

Uploads a file and attaches it to a Taskgert, typically the proof that its target was met: a screenshot (PNG, JPEG, WebP) or a recording (WebM, MP4) of the finished work. Give the content as base64. For large files from a terminal, the CLI reads the arguments from a file (`@args.json`) or stdin (`-`).

Abandons a Taskgert that is not finished. Its tasks are left as they are.

Marks a checklist item as done (or not done). Identify it by id or by its exact text. Needs tasks:write.

Marks several checklist items as done (or not done) in one call, on one or several tasks: e.g. "Release with gitflow and push", "Deploy on Coolify" and "Verify in production through the MCP" on every task of a run. Items are given by id or exact text; those a task does not have are listed in the answer and do not stop the rest. Needs tasks:write.

Picks an option of a task's decision (by label). An agent settling an unanswered, NON-critical decision passes asRecommendedDefault with the recommended option, and comments that it did; critical ones are answered by people (ask them). option null takes an answer back. Needs tasks:write.

In a running Taskgert, gives you (an agent, by your label) the next task you may work on while other agents work on theirs: never one another agent holds, nor one sharing an area with work in progress. Keep it alive with heartbeat_claim, and end it with release_task. Answers your current claim if you already hold one, or why nothing can be taken now (wait and ask again). With the run's autoMove on (the default), a task still in the first column is moved into the first active column, so its worked time counts from now. Refused once the run's agent limit or token budget is reached.

Saves a named, ordered set of checks in the project's vocabulary, e.g. "Standard" or "Bug fix". Propose it to the user first when it comes from patterns you noticed. Needs tasks:write and an editor role.

Creates an epic: a group of related tasks of a project. Put tasks in it with create_task or update_task (`epic`); a whole epic can be added to a Taskgert by its key. Needs tasks:write.

Creates a task in a project, at the bottom of a column (the first column by default). Needs the tasks:write scope.

Creates a Taskgert in a project: a batch of tasks with a target, run in order. It starts as a DRAFT. A task can belong to one Taskgert at a time.

Closes a running Taskgert once every task is done and its target has been carried out: records whether the target was met and a report of what was checked. Refused while tasks remain.

Returns a project's kanban board: its columns in order, each with its tasks in order. The best overview of a project.

The project's own words for verifying work: its checklist templates (with how to verify each check), every check its tasks use (most used first), the `Fix:` items (what went wrong while working), and suggestions of checks to add to a template. Read it before giving a task its checklist, and after a task or a Taskgert to propose refinements to the user.

Reads this bot's MCP calls over the last days: per tool, calls, errors, latency (p50/p95), failures retried and the longest run of back-to-back calls, with suggestions in plain words. The self-improvement loop: read it after a task or a Taskgert, propose improvement tasks to the user, ship them, and read it again to see whether they helped.

Lists this bot's recent tool calls through the MCP, newest first: tool, inputs (redacted), how they went, the first line of the answer and how long they took. Use it to see where an agent got stuck: errors, retries, long runs of the same tool. Kept 30 days.

Returns a task in full: description (Markdown), checklist, assignees, labels, attachments and, unless disabled, its comments. Read an attachment with read_attachment.

Returns a Taskgert in full: target, plan, status and phase, its tasks in plan order (column, complexity, worked and cycle time), the next task, and its time metrics. Start here to run one.

Extends your claim's lease by as long as it was first given. Call it while working (e.g. every few minutes). An expired claim cannot be renewed: its task went back to the pool, so claim again.

Records that `task` depends on `dependsOn`: it should only start once that one is done, and a running Taskgert will not hand it out before. Both must be in the same project; a link that would make a loop is refused. Needs tasks:write.

Lists the columns of a project in board order. Tasks in a column marked isDone count as completed.

Lists a project's epics (groups of related tasks, keyed like TGT-E1) with how many of their tasks are done. Their tasks: list_tasks with `epic`.

Lists the projects this bot was added to, with its role in each. Use the key (e.g. "TGT") to refer to a project in other tools.

Runs people asked for from Telegram (`/run`, with an optional note), newest first. A PENDING one is yours to answer: look at the open tasks of the project (list_tasks, completed false; Ideas are not runnable), judge whether they make sense as one Taskgert or several (dependencies, shared areas, risk), create and plan them (create_taskgert, plan_taskgert), then propose_run. PROPOSED: waiting for the person; APPROVED: start them (start_taskgert) and run them; REJECTED: their Taskgerts were cancelled.

Lists a project's Taskgerts (batches of tasks with a target, run in order), newest first, with status and progress.

Lists the tasks of a project, optionally filtered. Returns at most `limit` tasks and a cursor for more.

Moves a task to a column (possibly the same one) at the top, the bottom, or next to another task. Moving into a "done" column completes it. Needs tasks:write.

Records an agreed plan: the tasks in the order to run them (each with an optional complexity), the reasoning, and the target if it has none. A DRAFT becomes PLANNED. Propose the plan to the user and get their approval before calling this.

Answers a PENDING run request with the Taskgerts you created and planned for it (not started), in the order to run them, and why you grouped the tasks so. The person who asked gets it on Telegram with Approve / Reject buttons; check back with list_run_requests. On APPROVED, start them; on REJECTED they are cancelled for you.

Reads a file attached to a task or to one of its comments: a fresh download link and, unless disabled, its content — text for text, Markdown, CSV and JSON files, the picture itself for PNG, JPEG, GIF and WebP images. Other files (PDF, ZIP, Office) only get the link. Refer to the file by id, or by its name within the task.

Ends your work on a claimed task: DONE once it is delivered (with autoMove on, the run moves the task to its done column; otherwise move it yourself), or ABANDONED to hand it back. Report what you used, so the run can account for tokens and time. Then claim_next_task again.

Takes tasks out of a Taskgert's plan (the tasks themselves are untouched). While it runs, completed tasks stay.

Searches every project this bot can see, by text in the title or by exact key (e.g. "TGT-12").

Starts running a Taskgert (it needs a target and at least one task): the clock starts. Then work its tasks in plan order, moving each across the board as it progresses.

Undoes link_tasks: `task` no longer depends on `dependsOn`. Needs tasks:write.

Refines a template: rename it, replace its checks (e.g. add one the vocabulary suggests), or change which tasks it is for. Tasks that already have its checks keep them. Needs tasks:write and an editor role.

Changes an epic's name or description. Needs tasks:write.

Changes a task's title, description, priority, type, epic, due date or labels. Only the fields given change. Refining an Idea ends with type TASK. Pass `version` (from get_task) to fail instead of overwriting someone else's concurrent change.

Uploads a file and attaches it to a task. Give the content as base64 (`contentBase64`) or, for text files, as plain text (`text`). Allowed: images (PNG, JPEG, GIF, WebP), PDF, ZIP, Office documents, plain text, Markdown, CSV, JSON. Needs attachments:write and tasks:write.

Describes the bot this token belongs to, what the token allows (its scopes) and this server: its version and how many tools it has. Call it first. If your client lists fewer taskgert tools than `server.tools`, its tool list is out of date: ask the user to reconnect the MCP.