@dalmasonto/taskflow-mcp

Confirm which TaskFlow agent identity and project this credential maps to, plus the connection and terminal-mirror state. Connection and heartbeat are automatic — this confirms them, it does not establish them. mirror.state

Choose which agent identity this terminal is, when the repo defines several. Call this ONLY after asking your human which one to use — never guess. The choice is remembered for this terminal, so you will not be asked again after a reconnect.

List tasks in the agent

List the chat channels this agent can see in its project (shared rooms plus any DMs it is on the roster of).

List the other agents in this project so you know who to address.

Create a task in the agent

Edit a task

Advance a task

Self-assign a task in this agent

Record a review verdict on a task (decision: approved | changes_requested). Reports back to the assigned agent and transitions the task. Consider the

Send a chat message as this agent into a channel. Use list_channels to find channel ids.

Download a message attachment to disk and return its path. Get the `url` from an attachment on a message returned by check_messages. Returns a PATH, not the file

Check for messages addressed to you. Returns only what you have NOT yet marked read, across every channel you

Mark how far this agent has read in a channel (advance the read cursor forward).

Register (or reconnect) a live session so humans see this agent online. Defaults the identifier to host:pid and cwd to the current directory.

Send a liveness heartbeat for the current session (auto-registers one if needed). Optional status hint: idle | busy.

Stream a chunk of terminal output into the current session (auto-registers one if needed). Humans can watch it live.

Log a real activity event (e.g. an action you took). Optionally attach it to a task.

Read recent activity in the project (newest first). Optional task filter and limit.

Get a high-level analytics summary: total focused time, task completion rates, status distribution, and time per project. Useful for standup reports or understanding workload.

Get focused time grouped by day or week. Use for visualizing work patterns over time.

Get tool execution statistics: call count, success rate, average duration per tool. Shows which tools are used most and which are slow or failing.

Get per-task cost metrics: tool calls made during active timer sessions, total execution time, and tool breakdown. Useful for understanding which tasks consume the most resources.

Register this agent with a custom name. Optional — agents auto-register on startup using the project folder name. Call this only if you want a specific name.

Delete a stale or disconnected agent entry from the registry. Only disconnected agents can be deleted. Use this to clean up ghost entries left by crashed or renamed sessions.

Post a question to the TaskFlow Agent Inbox for the user to answer remotely. Returns immediately with the message ID. The question appears in the Agent Inbox UI with full context and optional quick-tap choices. After posting, use check_response to retrieve the user\

Check if a previously posted question (via ask_user or ask_agent) has been answered. Returns the response if answered, or status

Send a message to another agent by name. Returns immediately. The recipient agent will receive the message in their terminal (if running in tmux).

Ask another agent a question and wait for their response. Like ask_user but targets an agent. Returns the message ID — use check_response to poll for the answer. The recipient agent receives the question in their terminal (if in tmux) and can respond with respond_to_message.

Respond to a pending message addressed to this agent. Use check_messages to see incoming messages, then respond by message ID.

Send a question to multiple agents simultaneously. Creates a group message — each agent gets their own copy linked by a shared broadcast ID. Use check_broadcast to see all responses. Optional: omit agents list to broadcast to ALL connected agents.

Check the status of a broadcast message — shows which agents have responded and their answers.

List notifications. Check with unread_only=true at conversation start to surface important updates for the user.

Mark a notification as read after surfacing it to the user.

Mark all unread notifications as read. Call after the user has been briefed on pending notifications.

Delete all notifications. Use with caution — this is irreversible.

Start a timer session for a task. Call this before beginning work on any task to track focused time. Automatically transitions the task to

Pause the active timer session for a task. Call when switching context or waiting for user input. The task transitions to

Stop the active timer and set a final status. Call when finishing work — use

List timer sessions with optional filters. Use to review time spent on a task or across a date range.

Send raw keystrokes to an agent\

**Call this at the start of every conversation.** Returns onboarding instructions, behavioral rules, and live project context for AI agents working with TaskFlow. This tool tells you how to proactively manage tasks, track time, and stay in sync with the project.

**CALL THIS FIRST.** One-shot startup: returns your agent instructions, active project (auto-detected from folder name), open/blocked tasks, unread notifications, and registered agents — all in one call. Replaces the need to call get_agent_instructions + search_projects + list_tasks + list_notifications + list_agents separately.

Delete ALL tasks, projects, sessions, notifications, and activity logs. Settings are preserved. Use with extreme caution — this is irreversible.

Create a checkpoint snapshot of a task\

Get the latest checkpoint for a task. Returns the full task snapshot — useful for resuming work after an interruption.

List all available checkpoints for a task.

Get a setting value by key. Server settings (port, host, databasePath, logLevel, agentLivenessInterval, maxPortAttempts) come from ~/.taskflow_config.json. UI settings (timerBarDisplayMode, notificationInterval, statusColors, operatorName, systemName, etc.) come from SQLite.

Update a setting value. Server settings (port, host, databasePath, logLevel, agentLivenessInterval, maxPortAttempts) are written to ~/.taskflow_config.json and require a restart. UI settings are saved to the local SQLite database.

Retrieve recent activity log entries. Shows what has changed — task completions, timer events, status transitions. Filter by action or entity type.

Delete all activity log entries. Use with caution — this is irreversible.

Log a debug entry to the activity log. Use this to record your work process — what you investigated, commands you ran, decisions you made, and findings. Entries appear in the Activity Pulse and on the project page. Link to a task (task_id) or project (project_id) for context.

Compact old activity log entries into a summary. Keeps the most recent entries (default 50) and summarizes the rest. Use dry_run=true to preview what would be compacted. Runs automatically when log exceeds max_entries threshold.

Create a new project. Projects group related tasks and track time across them.

List all projects with task count. Call this at conversation start to understand the workspace.

Get a project by ID with all its tasks. Use this to understand the full scope of a project before starting work.

Update project fields such as name, color, type, or description.

Delete a project by ID. Tasks under this project will be unlinked (project_id set to NULL), not deleted.

Search projects by name or description. Use this to find projects related to your current work.

Get a task by ID with time tracking info. Read the description carefully — it often contains implementation details and acceptance criteria.

Delete a task by ID. Use sparingly — prefer updating status to

Create multiple tasks in a single transaction. Useful when breaking down a feature into subtasks.

Search tasks by title or description. Use this to find tasks related to your current work before creating duplicates.