Agents

MCP tools

Atlas groups its local stdio tools by consequence. The default profile can inspect work and plan changes. The workflow profile adds the full ticket loop agents need day to day. Delivery and admin stay separate because they reach provider, sync, archive, import, compact, or worktree state.

Your installed binary is the authority. Run tracker mcp tools --json --tool-profile workflow to inspect the exact names and schemas available in that build. This page explains the current source in useful groups.

Choose a profile by job

ProfileExposesUse it for
readlive inventoryInspection, queues, plans, context, status, attention, workspace list, backup health, and dry runs. Default.
workflowlive inventoryRead plus the ticket loop, workspace lifecycle, views, local checkpoint run, and restore plan.
deliverylive inventoryWorkflow plus dispatch and provider-backed change/check synchronization, with guarded delivery actions for supervised sessions.
adminlive inventoryHigh-impact operations including backup configure, restore apply, and workspace fork-copy. Each call also needs an external one-time approval.

Profiles are cumulative. Selecting workflow includes every read tool; selecting delivery includes read and workflow. Admin tools remain hidden unless the server also starts with --dangerously-allow-high-impact-tools.

Read tools

The read profile answers “what is true?” and “what would this operation do?” without mutating workflow state.

Queues and tickets

Find actionable work, inspect policy and dependencies, and read ticket history.

  • atlas.queue
  • atlas.next
  • atlas.search
  • atlas.board
  • atlas.ticket.view
  • atlas.ticket.history
  • atlas.ticket.inspect
  • atlas.context
  • atlas.status
  • atlas.backup.status
  • atlas.attention
  • atlas.activity
  • atlas.project.list
  • atlas.workspace.list
  • atlas.workspace.get
  • atlas.workspace.status
  • atlas.workspace.board_url

Agents and teams

Read what an agent can do now, why it is waiting, its profile, team presets, and wake-ups.

  • atlas.agent.available
  • atlas.agent.pending
  • atlas.agent.list
  • atlas.agent.view
  • atlas.agent.wakeup.list
  • atlas.agent.wakeup.view
  • atlas.team.list
  • atlas.team.show
  • atlas.goal.brief

Runs and review

Follow schedules, delivery state, evidence, handoffs, approvals, and the human inbox.

  • atlas.schedule.list
  • atlas.schedule.history
  • atlas.dashboard
  • atlas.timeline
  • atlas.run.view
  • atlas.evidence.list
  • atlas.evidence.view
  • atlas.handoff.view
  • atlas.approvals
  • atlas.inbox

Delivery state and plans

Inspect changes, checks, sync, conflicts, archive state, and guarded operation plans.

  • atlas.change.status
  • atlas.checks.list
  • atlas.sync.status
  • atlas.conflict.list
  • atlas.conflict.view
  • atlas.archive.plan
  • atlas.dispatch.suggest
  • atlas.dispatch.plan
  • atlas.change.merge_plan
  • atlas.sync.pull_plan
  • atlas.bundle.import_plan
  • atlas.archive.apply_plan
  • atlas.import.apply_plan
  • atlas.compact_plan
  • atlas.worktree.cleanup_plan

Workflow tools

This profile is the practical default when an agent is expected to do ticket work. Ticket and event mutations require actor and reason; they still run the same service permissions, dependency rules, leases, write lock, and event recording as the CLI. atlas.project.create is the narrow exception: it accepts only key and name, creates a metadata container, and does not write an attributed event. Run tracker mcp tools --json --tool-profile workflow for the live list.

Create and organize

Create project containers and tickets, edit ordinary ticket fields, set ownership and priority, manage labels, and build dependency or parent relationships without leaving MCP.

  • atlas.project.create
  • atlas.project.update
  • atlas.ticket.create
  • atlas.ticket.edit
  • atlas.ticket.assign
  • atlas.ticket.priority
  • atlas.ticket.label.add
  • atlas.ticket.label.remove
  • atlas.ticket.link
  • atlas.ticket.unlink
  • atlas.workspace.init
  • atlas.workspace.register
  • atlas.workspace.repair
  • atlas.workspace.hide
  • atlas.workspace.unhide
  • atlas.workspace.remove_pointer
  • atlas.views.save
  • atlas.views.delete
  • atlas.ticket.bulk
  • atlas.backup.run
  • atlas.restore.plan
  • atlas.settings.update

Work the ticket

Acquire, renew, and release leases, move through legal states, and leave durable progress notes.

  • atlas.ticket.claim
  • atlas.ticket.heartbeat
  • atlas.ticket.release
  • atlas.ticket.move
  • atlas.ticket.comment

Review and complete

Request review, approve or reject it, complete open-mode or separately gated work, and decide normal approval gates. Under review_gate, approval itself moves the ticket to Done.

  • atlas.ticket.request_review
  • atlas.ticket.approve
  • atlas.ticket.reject
  • atlas.ticket.complete
  • atlas.gate.approve
  • atlas.gate.reject

Configure agents

Create and edit profiles, enable or disable them, apply a preset, and acknowledge work wake-ups.

  • atlas.agent.create
  • atlas.agent.edit
  • atlas.agent.enable
  • atlas.agent.disable
  • atlas.team.apply
  • atlas.agent.wakeup.ack

Schedule and prove

Set one-time schedules, record checkpoints and evidence, and create the handoff for the next actor.

  • atlas.schedule.set
  • atlas.schedule.clear
  • atlas.run.checkpoint
  • atlas.evidence.add
  • atlas.handoff.create

Preview imports

Create a local import preview job before any high-impact apply step.

  • atlas.import.preview

atlas.ticket.edit exposes title, description, acceptance criteria, priority, labels, assignee, and reviewer. Omitted fields remain unchanged; empty descriptions, actor strings, and arrays clear their fields. Description whitespace reaches the service unchanged, while the markdown storage codec applies its normal surrounding-whitespace normalization when it persists the ticket.

Delivery and high-impact tools

The four delivery tools can create provider or runtime side effects: atlas.dispatch.run, atlas.change.create, atlas.change.sync, and atlas.checks.sync.

High-impact tools stay hidden unless the server starts with --dangerously-allow-high-impact-tools and the call carries a one-time approval. They cover provider review/merge, gate waiver, sync push/pull, bundle/import apply, archive apply/restore, compact, worktree cleanup, plus v1.15 atlas.backup.configure, atlas.restore.apply, atlas.workspace.fork_copy, and atlas.settings.grant_discovery. They are not in the Atlas-managed global registration. See MCP security for the external approval handshake and exact-target rules. Run tracker mcp tools --json --tool-profile admin --dangerously-allow-high-impact-tools for the live list.

A practical first loop

workflow profile
1. atlas.agent.available  { "agent_id": "builder-1" }
2. atlas.ticket.claim    { "ticket_id": "APP-12", "actor": "agent:builder-1", "reason": "start work" }
3. atlas.ticket.move     { "ticket_id": "APP-12", "status": "in_progress", "actor": "agent:builder-1", "reason": "start work" }
4. atlas.ticket.comment  { "ticket_id": "APP-12", "body": "implemented retry policy", "actor": "agent:builder-1", "reason": "progress" }
5. atlas.ticket.request_review { "ticket_id": "APP-12", "actor": "agent:builder-1", "reason": "ready for review" }

Argument schemas are discoverable from the server. Run tracker mcp schema --json --tool-profile workflow rather than copying guessed fields into production calls.