// SCAN REPORT

ATMcp

29 tools. 20 can modify or destroy data without limits.

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

Last updated: 2 Jul 2026

20 can modify or destroy data

9 read-only

29 tools total

Community server · catalogue entry verified 02/07/2026

How to control ATMcp ↓

TOOL MAP

What ATMcp exposes to your agents

Read (9) Write / Execute (18) Destructive / Financial (2)

Running more than one server? Check your whole stack →

MOST DANGEROUS TOOLS

Critical Risk

The most dangerous ATMcp tools

Destructive · Medium cancel_directive Cancelling a directive is an irreversible action — once cancelled, the directive's in-progress state and any associated work cannot be resumed or undone. Destructive · Medium retract_knowledge Tombstoning a knowledge entry is a removal/deletion operation. Execute · Medium wait_directive While 'wait' might initially suggest a passive Read operation, the description reveals this tool actively polls and waits on external directives in a distributed system context. Execute · High send_directive Sending a command to another agent causes that agent to execute actions whose effects depend entirely on the command's content.

20 of ATMcp's 29 tools can modify, destroy, or commit something on every call — and an agent calls them with no built-in limits.

POLICY

How to control ATMcp

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

Deny destructive operations

{
  "cancel_directive": {
    "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

{
  "append_output": {
    "limits": [
      {
        "counter": "append_output_per_hour",
        "window": "hour",
        "max": 30,
        "scope": "grant"
      }
    ]
  }
}

Prevents bulk unintended modifications from agents caught in loops.

Cap read operations

{
  "get_agent_output": {
    "limits": [
      {
        "counter": "get_agent_output_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 ATMcp — 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 ATMCP →

Instant setup, no code required.

FULL CATALOGUE

All 29 ATMcp tools

DESTRUCTIVE 2 tools

Destructive cancel_directive Cancel a directive you issued (only if it hasn't finished). Destructive retract_knowledge Tombstone a knowledge entry (OR-Set remove). A concurrent re-post wins (add-bias).

EXECUTE 2 tools

Execute send_directive Send a command to ONE specific agent (by display_name or agent_id). The target picks Execute wait_directive Await a directive you sent. With wait_ms > 0, long-polls until it reaches a terminal

WRITE 16 tools

Write append_output Stream a chunk of YOUR output/progress so the console can watch it live (optionally Write claim_directive Mark a directive addressed to you as 'running' before you start executing it. Write claim_next_task Claim the highest-priority eligible (deps satisfied) open task in one round-trip. Write claim_task Atomically claim a specific open task. Returns a fencing_token you must present Write complete_task Mark a task done. Returns eligible_unblocked — downstream tasks now claimable. Write create_goal Create a team goal that tasks can roll up into. Write create_task Create a task. depends_on lists task_ids that must be done first; weight Write fail_task Report a task failed. Re-queued (open) until max_attempts, then dead-lettered (failed). Write heartbeat Refresh your presence (~every 10s). Also renews any task lease you hold and Write join_team Join (or re-join) a team. The join token may be passed here or sent as an Write leave_team Gracefully leave the team: releases held tasks and marks you offline. Write post_knowledge Share a finding. Identical content (same title/body/tags) is auto-deduped; Write release_task Voluntarily give a task back to the queue (instant re-queue, no lease wait). Write report_directive Report a directive finished: status must be 'done' or 'failed'. The issuer's Write set_memory Set a shared-memory key (LWW by team logical clock). Pass expected_version Write update_task_progress Report progress and renew your lease. status may be 'in_progress' or 'blocked'.

READ 9 tools

Read get_agent_output Tail another agent's output (by display_name or agent_id) from since_seq. With Read get_memory Read a shared-memory key, or all keys if key is omitted. Read get_team_status One-shot snapshot: agents online, task counts, goal progress %, knowledge count, head cursor. Read inbox Your directive inbox: pending commands addressed to you. With wait_ms > 0 this Read list_agents Team roster with derived presence (healthy/degraded/offline), status, and progress. Read list_directives List directives. role='sent' (issued by you), 'received' (addressed to you), or omit Read list_tasks List the task board, optionally filtered by status/assignee/goal or since a cursor. Read search_knowledge Full-text search the team knowledge base (FTS5). Omit query to list recent Read sync Catch up on events since a cursor. If nothing new and wait_ms > 0, long-poll up to

EXPLORE

Related servers

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

FAQ

Questions about ATMcp

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

Yes. The ATMcp server exposes 2 destructive tools including cancel_directive, retract_knowledge. 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 ATMcp? +

The ATMcp server has 16 write tools including append_output, claim_directive, claim_next_task. 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 ATMcp.

How many tools does the AT MCP server expose? +

29 tools across 4 categories: Destructive, Execute, Read, Write. 9 are read-only. 20 can modify, create, or delete data.

How do I enforce a policy on ATMcp? +

Register the AT 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 ATMcp tool call.

Deterministic rules across all 29 ATMcp tools. Per-identity grants. Full audit log. Live in minutes. Nothing to install.

ENFORCE POLICY ON ATMCP →

Instant setup, no code required.

29 ATMcp tools catalogued and risk-classified — across an index of 46,500+ MCP servers.

// WHERE THIS COMES FROM

These policies come from AT's registry record.

The record behind this page: verified identity, auth posture, risk grade, every tool classified, recommended policy — re-checked continuously.

Teams ship this data inside their own products. See what a licence covers →