// SCAN REPORT

Agent Bus

65 tools. 34 can modify or destroy data without limits.

3 destructive tools with no built-in limits. Policy required.

Last updated: 28 Jun 2026

34 can modify or destroy data

31 read-only

65 tools total

Community server · catalogue entry verified 28/06/2026

How to control Agent Bus ↓

TOOL MAP

What Agent Bus exposes to your agents

Read (31) Write / Execute (31) Destructive / Financial (3)

Running more than one server? Check your whole stack →

MOST DANGEROUS TOOLS

Critical Risk

The most dangerous Agent Bus tools

Destructive · High cancel_task Cancelling a task is an irreversible action — once cancelled, the in-progress work and task state are terminated and cannot be undone. Destructive · High delete_team This tool permanently removes team membership records and erases team labels from history. Destructive · Medium remove_agent Removing an agent from the live roster is an irreversible action that affects system state — the agent is taken off the active roster. Execute · High wait_for_task While the tool itself is read-like (retrieving task state), it functions as an execution control point in the agent-bus architecture.

34 of Agent Bus's 65 tools can modify, destroy, or commit something on every call — and an agent calls them with no built-in limits.

POLICY

How to control Agent Bus

PolicyLayer is an MCP gateway — it sits between your AI agents and Agent Bus, and nothing reaches the server without passing your rules. These are the rules we recommend:

Deny destructive operations

{
  "cancel_task": {
    "deny_if": [
      {
        "conditions": [],
        "on_deny": "Blocked by default. Requires approval."
      }
    ]
  }
}

Destructive tools should never be available to autonomous agents without human approval.

Rate limit write operations

{
  "ack": {
    "limits": [
      {
        "counter": "ack_per_hour",
        "window": "hour",
        "max": 30,
        "scope": "grant"
      }
    ]
  }
}

Prevents bulk unintended modifications from agents caught in loops.

Cap read operations

{
  "activity": {
    "limits": [
      {
        "counter": "activity_per_minute",
        "window": "minute",
        "max": 60,
        "scope": "grant"
      }
    ]
  }
}

Controls API costs and prevents retry loops from exhausting upstream rate limits.

Create a free account and register Agent Bus — nothing to install.
Add these rules — paste them, or build them visually. Tune the limits to your setup.
Point your MCP client (Claude, Cursor, anything) at your gateway URL.

ENFORCE POLICY ON AGENT BUS →

Instant setup, no code required.

FULL CATALOGUE

All 65 Agent Bus tools

DESTRUCTIVE 3 tools

Destructive cancel_task Cancel a non-terminal task, record a cancel event, notify the other side, and run task.canceled hooks. Destructive delete_team Delete a team scope by unregistering its live members and clearing that team label from preserved history. Destructive remove_agent Remove an agent/member from the live roster while preserving message and task history.

EXECUTE 3 tools

Execute ask Send a question and BLOCK until a reply arrives (or timeout). Capped at 110s. Execute ask_best Route an ask to the most recently-active agent that has the given capability. Execute wait_for_task Wait up to 110s for a task update/event/message/test result, then return task_result plus holder, latest activ

WRITE 28 tools

Write ack Acknowledge a message you successfully processed (used with inbox claim_s). Write acknowledge_task Acknowledge an assigned task as claimed, declined, or blocked. Sends a receipt back to the requester. Write ask_async Create an ask message and return immediately with ask_id, recipient presence, and next actions. Write assign_task Assign an open task directly to an agent, moving it to claimed. Write claim_task Atomically claim an Write create_task Create a new task in state Write delegate Create a task, assign it to an agent, require acknowledgement by default, notify the assignee, and record a de Write delegate_team Create board-visible tracked tasks for active members of a team. Write handoff_task Create a pinned handoff memory for a task and optionally assign it to another agent, or release it if no targe Write now Update an agent Write pin_memory Pin a memory so session_brief surfaces it in the pinned handoff section. Write record_decision Persist a project decision with rationale and implemented flag. Write record_task_event Append a task progress/event log row. Use for phases, progress notes, command summaries, result notes, and can Write record_test_result Record explicit test/build/lint evidence for final_report. Write register Register this session as an agent on the bus. Pick a stable name like Write release_task Return a held task to Write remember Persist a structured memory for future sessions: summary, handoff, risk, todo, fact, blocker, lesson, gotcha, Write reply Answer a pending ask. The original ask Write reply_thread Continue an existing thread without remembering the exact recipient. Sends to the last other participant in th Write send Send a fire-and-forget message to another agent Write send_channel Broadcast a message to every subscriber of a channel. Write send_team Send one message to every active agent in a team, optionally scoped by project/area. Use for private team fan- Write set_agent_status Set an agent work state: idle, working, blocked, waiting_review, or sleeping. Write submit_review Submit verifier review for a task. Approved reviews satisfy review gates; rejected reviews mark changes_reques Write unpin_memory Unpin a memory so it returns to normal recent-memory ordering. Write unsubscribe Remove this agent from a channel. Write update_task Update a task. State transitions are strict: backlog->open|canceled, open->backlog|claimed|canceled, claimed-> Write wake_agent Wake a sleeping agent by setting status to idle.

READ 31 tools

Read activity Chronological project/area/team activity timeline across messages, task events, test results, decisions, and m Read ask_team Ask the best active member of a team. Optionally filter by capability and role. Use for team-scoped request/re Read check_scope_conflicts Check whether a proposed file_scope overlaps active claimed/working/blocked tasks in the same project/area. Read claim_best_task Claim the highest-priority open task in this agent Read cockpit Coordinator dashboard: waiting items, ready items, blockers, suggested next actions, and the underlying projec Read directory List registered agents with status, age, scope, role, capabilities, and active task id. Read final_report Generate a merge-readiness report from tasks. Read get_message Fetch one exact message by id. Use include_content:false or preview_chars to avoid returning a huge body. Read get_task Fetch a single task by id. Read inbox Read new messages addressed to you. Read inbox_previews Preview pending inbox messages without consuming them and without returning full message bodies. Read inbox_status Inspect an agent inbox without consuming messages. Shows unread, claimed/in-flight, recent delivered messages, Read list_decisions List persisted decisions for the current project/area. Read list_memories List structured memories for the current project/area, optionally filtered by agent, kind, task, or thread. Read list_task_events List task event/progress rows, optionally filtered by task, agent, type, project, or area. Read list_tasks List tasks. By default excludes terminal states (completed/failed/canceled) but includes backlog; set include_ Read list_test_results List recorded test/build/lint evidence. Read message_status Diagnose one message: delivery/claim state, reply, recipient presence, related task, and suggested next action Read project_board Manager board: agents, open/active/blocked/waiting-review/stale tasks, scope conflicts, pinned risks/handoffs, Read recent Read the most recent messages on the bus regardless of recipient. Useful for catching up. Read review_gate Return a deterministic merge/push gate from the project board and final report. ok=false includes blockers and Read session_brief Generate a startup/handoff brief from live agents, open/blocked/stale tasks, recent decisions, memories, and m Read sleep_agent Mark an agent as sleeping. Read subscribers List the agents subscribed to a channel. Read task_result Fetch one task plus its event log, test evidence, related memories, and thread messages. Read team_board Manager board for one team: agents, tasks, review queue, acknowledgements, conflicts, pinned risks/handoffs, a Read thread Read all messages in a conversation thread, in order. Read wait_for_agents Wait for an expected roster of agents and report ready, missing, stale, and wrong-scope registrations. Read whois List every agent currently registered on the bus along with capabilities and last-seen timestamps. Read why_no_reply Explain why an ask/message has no reply yet, including recipient presence, claim state, and related task conte

EXPLORE

Related servers

Other MCP servers with similar tools — same risk classification, starter policies for each.

FAQ

Questions about Agent Bus

Can an AI agent delete data through the Agent Bus MCP server? +

Yes. The Agent Bus server exposes 3 destructive tools including cancel_task, delete_team, remove_agent. These permanently remove resources with no undo. PolicyLayer blocks destructive tools by default so they never reach the upstream server.

How do I prevent bulk modifications through Agent Bus? +

The Agent Bus server has 28 write tools including ack, acknowledge_task, ask_async. Set a rate limit in your policy -- for example, 10 calls per hour prevents an agent from making more than 10 modifications per hour. PolicyLayer enforces this at the gateway, before calls reach Agent Bus.

How many tools does the Agent Bus MCP server expose? +

65 tools across 4 categories: Destructive, Execute, Read, Write. 31 are read-only. 34 can modify, create, or delete data.

How do I enforce a policy on Agent Bus? +

Register the Agent Bus MCP server in PolicyLayer, apply the suggested rules above (adjust the limits to your use case), and point your AI client at the PolicyLayer proxy URL instead of the server directly. Your agents keep the same tools; PolicyLayer evaluates every call against policy before it executes. Nothing to install, live in minutes.

Enforce policy on every Agent Bus tool call.

Deterministic rules across all 65 Agent Bus tools. Per-identity grants. Full audit log. Live in minutes. Nothing to install.

ENFORCE POLICY ON AGENT BUS →

Instant setup, no code required.

65 Agent Bus tools catalogued and risk-classified — across an index of 43,000+ MCP servers.