New Your team’s decisions, in one playbook every coding agent works from. Never answer your agent twice

DutyDeck

247 tools. 118 can modify or destroy data without limits.

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

Last updated:

118 can modify or destroy data
129 read-only
247 tools total

Community server · catalogue entry checked 19/10/2026

How to control DutyDeck ↓

What DutyDeck exposes to your agents

Read (129) Write / Execute (108) Destructive / Financial (10)
Critical Risk

The most dangerous DutyDeck tools

118 of DutyDeck's 247 tools can modify, destroy, or commit something on every call — and an agent calls them with no built-in limits.

How to control DutyDeck

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

Deny destructive operations
{
  "canvas_cancel_media_run": {
    "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
{
  "api_bind_crypto_profile": {
    "limits": [
      {
        "counter": "api_bind_crypto_profile_per_hour",
        "window": "hour",
        "max": 30,
        "scope": "grant"
      }
    ]
  }
}

Prevents bulk unintended modifications from agents caught in loops.

Cap read operations
{
  "api_extract_base_url": {
    "limits": [
      {
        "counter": "api_extract_base_url_per_minute",
        "window": "minute",
        "max": 60,
        "scope": "grant"
      }
    ]
  }
}

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

  1. Create a free account and register DutyDeck — nothing to install.
  2. Add these rules — paste them, or build them visually. Tune the limits to your setup.
  3. Point your MCP client (Claude, Cursor, anything) at your gateway URL.
ENFORCE POLICY ON DUTYDECK →

Instant setup, no code required.

All 247 DutyDeck tools

DESTRUCTIVE 10 tools
Destructive canvas_cancel_media_run 取消确切音频或视频节点拥有的运行;必须显式声明取消意图。 Destructive factory_cancel_experiment Stop the experiment while preserving spent budget and completed evidence; cancellation does not refund quota. Destructive mcp__automation__delete_automation 删除 DutyDeck 定时任务。只在用户明确要求删除,或任务已经长期无价值且用户确认后使用。 Destructive mcp__planning__delete_calendar_event 删除日程。只在用户明确要求删除时使用;含 nativeOrigin 且来源为可写已连接系统日历时,会真实删除对应 macOS Calendar 日程,必须先说明该外部副作用并取得用户确认;只读来源会失败。仅 Pi Age Destructive mcp__planning__delete_group 删除指定范围内的分组,并仅清除该范围关联对象的分组字段。只在用户明确要求删除时使用。仅 Pi Agent 可用。 Destructive mcp__planning__delete_reminder 删除提醒记录。只在用户明确要求彻底删除提醒时使用。仅 Pi Agent 可用。 Destructive mcp__planning__delete_tag 删除标签并移除其关联。只在用户明确要求删除时使用。仅 Pi Agent 可用。 Destructive mcp__planning__delete_todo 删除 Todo。只在用户明确要求删除时使用;含 nativeOrigin 且来源为可写已连接系统提醒事项列表时,会真实删除对应 macOS Reminder,必须先说明该外部副作用并取得用户确认;只读来源会失败。仅 Pi Destructive media_cancel_generation 明确取消当前 Agent 会话持有的统一媒体任务;迟到响应不会复活已取消任务。 Destructive media_cancel_run 取消属于本次 Agent 的指定任务;必须明确声明取消意图,远端运行中任务不会调用全局中断。
EXECUTE 28 tools
Execute api_prepare_request Prepare a saved request or partial draft; resolve environment, runtime variables and overrides without sending Execute api_run_scenario Run a prepared scenario once. One approval authorizes every step of that flow: the approval card lists each st Execute BrowserExecuteJavaScript Run JavaScript in the current page context when fixed BrowserDomAction cannot achieve the user-requested task. Execute BrowserNavigate Navigate the Agent working in-app browser tab to a URL or search query. Explicit URLs, bare domains, localhost Execute BrowserNewTab Create a new Agent working tab and activate it in the visible in-app browser. Optionally navigate it to any UR Execute BrowserWaitFor Wait for a fixed page condition after navigation or an action: a URL fragment, visible text, or CSS selector. Execute canvas_run_agent 显式运行单个 Canvas Agent 并等待终态;不会自动运行下游节点或启动图片任务。专业审核用 reviewScope 明确整图或指定节点,返回 reviewCoverage 数据覆盖;首次导演规划可用 positi Execute canvas_run_nodes 运行已有图片、音频或视频节点并创建待验收候选,调用远端模型时可能产生费用;同批不能同时选择有直接依赖关系的上下游。 Execute canvas_run_workflow 从指定 Agent 起点运行一次 bound 可达下游;图片严格受本次上限约束并停在候选验收状态。 Execute canvas_task 执行前 start 登记交付;跨回合先 status 查看并 resume 原 taskId。created 是新节点,updated 是原节点新产物。读取真实证据后 complete;证据类提交只能是 {id,evid Execute factory_advance_experiment Run exactly the next split within the same 3 comparison rounds and 60 run units used by other tools. Only afte Execute factory_diagnose_run 只读诊断执行、约束和评审失败,不调用模型。返回源指纹、可补评步骤和已有独立补评记录。评审超时不代表业务输出错误;读取评审证据会登记关联实验暴露。 Execute factory_retry_review 先用 factory_diagnose_run 诊断,再绑定返回的 sourceFingerprint 执行一次有界补评。仅补 timeout/invalid-response 步骤,不重新生成业务输出,不改变原 run Execute factory_run_batch Run a persisted evaluation or baseline/candidate comparison batch over an existing dataset or saved task IDs. Execute factory_run_scene Run a whole scene without changing or adopting its definition. For an optimization comparison, run the current Execute factory_run_step Run ONE prompt step with the input you provide, then automatically assess its output against this run’s fixed Execute mcp__automation__run_automation_now 立即运行 DutyDeck 定时任务。用于用户要求马上验证,或修改任务后需要试跑一次。 Execute mcp__collaboration__stop_delegation 停止一个正在运行的 DutyDeck 协作子会话。 Execute mcp__collaboration__stop_delegations 批量停止多个正在运行的 DutyDeck 协作子会话。 Execute mcp__collaboration__wait_for_delegations 父会话用于收敛子会话结果的等待屏障:等待指定的 DutyDeck 协作子会话完成,并返回结构化结果摘要。只要本轮回复、决策或交付依赖已委派任务,主会话必须在回复前调用本工具,不能只因 delegate_agent/del Execute media_execute_generation 启动已准备的统一媒体生成任务;需要执行权限并遵守媒体自主策略。 Execute media_execute_run 启动已经准备且属于本次 Agent 的任务;上传、编译、提交与收集由统一监督器推进。 Execute media_wait_generation 等待统一媒体任务 revision 变化或终态;超时只返回当前快照,不取消远端任务。 Execute media_wait_run 等待任务 revision 变化或终态,最长 30 秒;停止等待不会取消远端任务。 Execute ops_database_change_context Read columns and indexes for up to four saved MySQL/PostgreSQL/SQLite tables outside the disabled-table list t Execute proma_knowledge_document Write one approved outline item. Prefer targetRef from status and pageRef/quote evidence from pages read in th Execute server_task_start 复用运维模块已连接的 SSH,在独立可见终端中执行发布或需要 sudo 交互的命令。每次启动都要单独确认;不受短命令 120 秒上限限制。密码只能由用户在运维任务终端输入,禁止写入命令或聊天。返回任务身份不代表执行成功。 Execute TerminalExecute Run one command in a visible Agent-owned terminal Tab. Prefer terminalId to reuse a safe matching current-sess
WRITE 80 tools
Write api_bind_crypto_profile Bind one crypto profile to existing requests with ONE approval (up to 50 requests per call). The approval card Write api_declare_variables Declare variable NAMES on the workspace (shared across all collections) or on one collection, with a separate Write api_save_crypto_profile Create or replace a signing/encryption profile with a separate configuration-write approval. A profile is a SH Write api_save_environment Create or update an environment (local/test/production) and its variables with a separate configuration-write Write api_save_request Save the prepared draft with a separate configuration-write approval. Use the preview catalogRevision as expec Write api_save_scenario Create or replace a scenario with a separate configuration-write approval. Steps only reference saved requests Write api_send_request Execute an approved preparedId once. Repeated calls reuse its run and never resend. Inspect errors before inte Write api_update_requests Bulk-edit configuration of existing requests with ONE approval: rename them to business-readable names, move t Write BrowserClick Click an element reference from the latest BrowserObserve result. References expire after navigation or a new Write BrowserClose Close every tab in the current in-app browser session and hide its browser panel. Write BrowserCloseTab Close a browser tab by tab id. Closing the last tab closes the in-app browser session. Write BrowserDrag Perform a native pointer drag from one current BrowserObserve/BrowserFind reference to another. It does not sy Write BrowserFill Replace all text in a referenced input, textarea, or contenteditable editor with complete text (including spac Write BrowserPress Press a navigation key (Enter, Tab, Escape, arrows, Backspace, Delete, etc.) or insert complete text into the Write BrowserPreviewOpen Open an HTML file or a directory containing index.html from the current project or an authorized attached dire Write BrowserScroll Scroll the document or an optional CSS-selected scroll container using a fixed, data-only operation. This repl Write BrowserSelectOption Select a native HTML <select> option by value, visible label, or zero-based index through a fixed DOM operatio Write BrowserSelectTab Switch the Agent working tab by tab id and activate that tab in the visible browser panel. Write BrowserUpload Set files on a current BrowserObserve/BrowserFind native file-input reference. Every path must be an absolute Write canvas_apply_changes 只通过受控批量事务提交画布结构修改;删除或覆盖必须声明 destructiveIntent=explicit,版本冲突时须重新读取并复核,不自动换用最新基线。新增或替换边使用:${CANVAS_UPSERT_EDGES_ Write canvas_attach_image_result 把冻结 attachment 的成功图片结果登记为原图片卡候选;配置已变化时保留旧基线并标记 stale,绝不自动覆盖或采用。 Write canvas_attach_media_assets 把当前会话通过 media_import_local_file 登记的精确资产回填为音视频节点候选。须先配置输出 key 并完整覆盖所有输出;不提交生成,也不自动采用。 Write canvas_attach_media_run 把当前普通 Agent 已成功的独立媒体运行显式登记为既有音频或视频节点候选;不会执行、取消或自动采用该运行。 Write canvas_create_agent 在已关联画布中创建独立 Canvas Agent 节点承担后续分工,可选从已有节点自动连线;普通 Agent 无需让用户手工创建。 Write canvas_create_artifact 在已关联画布中原子创建含真实内容的文档、WebView 原型或图片设计稿节点,可选从已有节点自动连线。不要把 Markdown、HTML 或图片提示词正文传给 canvas_apply_changes。 Write canvas_create_media 在已关联画布中原子创建音频或视频节点的空媒体模块;创建后用 canvas_update_media_config 保存 typed 输入输出,可选择已有预设或保留 profile=null 供工作流草稿试运行。 Write canvas_import_image 把当前 Agent 已授权目录中的现有图片导入已关联画布,并创建立即采用该图片的参考节点。仅导入图片时省略 sourceNodeId 和 relation;需要连线时两项必须同时提供,来源节点须位于当前画布。不要要求用户 Write canvas_manage 列出本会话已关联画布(包括归档),或创建、关联、解除关联、设置默认、改名、归档、恢复和删除整张画布。归档/恢复/改名/删除必须来自用户明确指令并传 intent=explicit;删除保留运行中保护,不能用于绕过未决任务 Write canvas_update_agent_config 局部更新已有 Canvas Agent 的长期职责、Skills 或模型选择;不会运行模型、下游节点或图片任务。 Write canvas_update_artifact 更新已有文档、WebView 正文或图片提示词;图片只保存配置,不会生图。用户要求立即生图时必须另行调用 canvas_run_nodes。初始正文 revision 0 可直接更新;版本冲突时先 canvas_read Write canvas_update_image_config 局部更新已有生图节点的提示词、模型、编辑底图、工作流、画幅或上下文;editSourceNodeId 只能选择通过 image.asset -> image.reference 绑定的图片节点,传 null 改用当前节点 Write canvas_update_media_config 以配置 revision 保存音视频卡片;workflow 固定公共或当前项目的工作流版本和连接,profile 仅兼容旧预设且两者互斥。先用 media_list_workflows/media_inspect_wor Write canvas_update_settings 仅在用户明确要求修改画布全局设置时使用,声明 settingsIntent=explicit,并传读取到的精确 baseRevision。可改媒体模型范围和新任务默认 ComfyUI 连接(null 清空)。版本冲突需重 Write EnterPlanMode 进入 DutyDeck 计划模式。进入后只能调研、整理计划,并等待用户批准后再执行写操作。 Write ExitPlanMode 向用户提交计划并请求批准。用户批准后才能退出计划模式并继续执行。 Write factory_apply_draft Step 2 of 2: write the previously prepared draft without changing the live version. Only a preparedId issued b Write factory_apply_operation Step 2 of 2: apply exactly one prepared operation after host authorization. Only a live preparedId is accepted Write factory_apply_stub Step 2 of 2: write the previously prepared stub. Only a preparedId issued by factory_prepare_stub is accepted, Write factory_create_experiment Freeze one prompt candidate, protocol, source-grouped splits and call/time limits without running models or ov Write factory_prepare_operation Step 1 of 2 for scene lifecycle, draft adoption/discard, rollback, export, dataset changes, or stub deletion. Write InstallWindowsShell Use this when the user task truly requires command execution but this Windows device has no Git Bash or WSL. I Write mcp__automation__create_automation 创建 DutyDeck 持久化定时任务。可通过 workspaceId 指定其他工作区,先用 list_workspaces 查询;省略则使用当前工作区。适合无人值守、有稳定价值的场景。纯提醒/闹钟、需要用户实时参与判断 Write mcp__automation__update_automation 修改 DutyDeck 定时任务,包括名称、执行提示词、频率和启用状态。定时任务自动执行中可以省略 id 来修改当前任务。 Write mcp__collaboration__set_delegation_thinking_level 父 Agent 修改自己创建的协作子会话思考强度。当前已运行的 turn 不会中途切换,新强度从下一轮或续跑开始生效。 Write mcp__planning__complete_todo 将指定 Todo 标记为已完成。若含 nativeOrigin 会同时完成用户已连接的系统提醒事项;必须先说明该外部副作用并取得用户确认。仅在任务确实完成或用户明确要求完成时使用。仅 Pi Agent 可用。 Write mcp__planning__create_calendar_event 创建 DutyDeck 本地日程。分组必须来自 list_groups({ scope: calendar });用户明确提供时间安排时使用。仅 Pi Agent 可用。 Write mcp__planning__create_group 创建 Todo 或日程范围内的独立分组。只在用户明确提出新分组或该范围内现有分组不适用时使用。仅 Pi Agent 可用。 Write mcp__planning__create_reminder 为 Todo 或日程创建指定时点的提醒。仅在用户要求提醒且时点明确时使用。仅 Pi Agent 可用。 Write mcp__planning__create_tag 创建跨 Todo 和日程复用的标签。只在用户明确给出新标签或现有标签不适用时使用。仅 Pi Agent 可用。 Write mcp__planning__create_todo 创建 DutyDeck 本地 Todo。调用前必须先用 list_todos(status=open) 检查重复,并用 list_groups({ scope: todo }) 查询并优先复用 Todo 分组;用户明确提 Write mcp__planning__update_calendar_event 更新日程时间或内容。若日程含 nativeOrigin,会写回用户已连接的系统日历;单项修改先确认,批量修改必须明确确认,只读来源会失败。仅 Pi Agent 可用。 Write mcp__planning__update_group 更新指定范围内的分组,不能借此移动分组范围。仅 Pi Agent 可用。 Write mcp__planning__update_reminder 修改未确认提醒的触发时间。仅 Pi Agent 可用。 Write mcp__planning__update_tag 更新标签名称或颜色。仅 Pi Agent 可用。 Write mcp__planning__update_todo 更新 Todo 的标题、说明、优先级或截止时间。若 Todo 含 nativeOrigin,此操作会写回用户已连接的系统提醒事项:单项编辑/完成先征得用户确认;批量修改必须明确确认;只读来源会失败。仅 Pi Agent Write media_import_assets 把当前会话已授权的不透明来源登记为项目媒体资产,不接受绝对路径。 Write media_import_local_file 把 Shell 或 Skill 在当前 Agent cwd、项目目录或本会话已授权附加目录生成的一个明确图片、音频或视频文件登记为项目媒体资产;不扫描目录,也不返回本地路径。 Write media_import_remote_asset 读取已发现的精确远端 asset 并登记为当前项目不可变素材;发现资源本身不会触发导入。 Write media_import_remote_workflow 兼容历史接入调用,当前等同于 media_use_remote_workflow:Host 自动复用远端执行快照,不创建项目草稿或本地模板。旧 id/name/revision 参数仅保留调用兼容,请始终使用回执返回的新 Write media_publish_workflow 在明确发布意图和执行权限下,把 API 工作流保存为所有项目可发现的不可变公共版本;不执行生成。 Write media_save_local_workflow 新任务先检查服务器工作流,无匹配时根据实际模型和节点新建。authorizationMode=ask 时先取得用户同意并传 creationIntent=user-confirmed;automatic 时可直接新建并传 Write media_save_profile 兼容旧流程,把连接与固定工作流版本保存为当前项目预设;新任务无需创建预设。 Write media_save_workflow_draft 仅兼容项目旧流程;新建旧草稿须传 creationIntent=user-confirmed 或在用户已开启自主模式时传 automatic-policy。新任务直接使用服务器工作流,无匹配时按当前生成授权策略处理,并改 Write ops_connection_prepare Prepare a new SSH, MySQL, PostgreSQL, Redis or remote SQLite connection draft for this user session. The Serve Write ops_database_write 对已保存的直连 MySQL、本地 SQLite 或已连接 SSH 主机上的 SQLite 文件执行有界写入脚本。每次执行前必须由用户在 Agent 原生确认弹窗中批准,完全自动模式也不能跳过,不需要额外的服务器 Agen Write ops_redis_write 对已保存 Redis 连接执行一次有界单键修改,每次必须由 Agent 原生确认弹窗批准,完全自动模式也不能跳过。支持 SET、DEL、UNLINK、EXPIRE、PEXPIRE、PERSIST、INCR、DECR、HS Write ops_server_discover Discover bounded systemd services and Docker containers on one connected saved server. Reports partial/unavail Write proma_knowledge_import_sources 用户要求导入旧知识库时打开原生文件或目录选择器;保留正文、附件、链接和来源版本,然后扫描。模型不得传入路径;取消选择不写入。根据导入回执继续检索、读取和智能归档。 Write proma_knowledge_outline Create or revise the readable document outline for the currently approved plan. Prefer flowRef from status, or Write proma_knowledge_plan Propose free-form knowledge groups from scanned source metadata. The proposal waits for user confirmation and Write proma_knowledge_submit Submit durable knowledge grounded in source pages actually read during this run or in an exact quote from the Write proma_workspace_configure_mcp_server Create or update a non-sensitive workspace MCP transport and validate it with a real handshake and listTools c Write SelectWorktree Bind this Agent session to an existing linked Git worktree that belongs to an authorized or registered reposit Write server_connect 检查已保存服务器的现有连接。此工具不会建立或重新连接 SSH;如需连接,请先在运维面板中连接服务器并确认主机密钥。 Write server_disconnect 断开一个已保存服务器的现有连接。此操作每次都需要单独确认;断开后,本轮 Agent 运行将不能继续使用该连接。 Write TaskCreate 创建一个可见进度任务,用于多步骤或长耗时工作。 Write TaskUpdate 更新已有可见进度任务的状态、标题或说明。 Write TerminalClose Close and terminate a terminal owned by the current Agent session. Write TerminalInterrupt Send Ctrl+C to a running terminal owned by the current Agent session. The terminal remains visible. Write TerminalOpen Open a visible terminal Tab in the Agent right workspace. cwd controls the initial directory and must resolve
READ 129 tools
Read api_extract_base_url Take the hardcoded host out of a collection\ Read api_get_request Read a saved request with secrets redacted. Use its ID with api_prepare_request. Read api_inspect_run Read bounded pages from an existing run in this session; never resend. section supports summary, request, head Read api_list List saved API requests and available environments in this session project. Does not contact servers. Read api_prepare_scenario Prepare a saved scenario (an ordered flow of saved requests) and return its step list: index, method and resol Read AskUserQuestion 当需要用户选择、补充信息或确认偏好时调用,DutyDeck 会展示可交互问答横幅。 Read BrowserAct Click a current BrowserObserve/BrowserFind reference and optionally wait for one URL, visible-text, or CSS-sel Read BrowserDomAction Use a CSS selector to focus, fill, click, or inspect a page element when BrowserObserve cannot locate a dynami Read BrowserExtract Extract compact text or basic Markdown from the document body or a CSS-selected region without arbitrary page Read BrowserFind Find fresh accessibility element references by semantic role and/or accessible name without returning a full p Read BrowserHover Move the native pointer over a current BrowserObserve/BrowserFind element reference. Use it to reveal hover me Read BrowserListTabs List all tabs in the current in-app browser session, including the user-visible tab and Agent working tab. Use Read BrowserObserve Read the current in-app browser URL, title, and compact accessibility snapshot. Each BrowserObserve or Browser Read BrowserScreenshot Capture the Agent working in-app browser page as a PNG. Use BrowserObserve first when semantic page structure Read canvas_adopt_image_candidates 采用查询得到的精确候选指纹;all 要求整批成功,succeeded 明确保留其余旧版本。只在用户已授权采用或自动推进的任务中调用,采用后可恢复原工作流。 Read canvas_adopt_media_candidate 按输出 key 原子采用指定媒体候选;同 bundle 输出必须一次完整选择,采用后持久工作流可继续推进。 Read canvas_arrange 复用前台画布自动整理,范围为选中、相关、可见、全部或指定节点。必须传读取到的baseRevision与intent=explicit;nodes范围必须给nodeIds,冲突不自动重试。只有当前普通主Agent可调用。 Read canvas_get_context 返回当前 Agent 已关联、默认、活动画布和本轮明确引用摘要;不会扫描项目全部画布。 Read canvas_get_image_candidates 查询精确批次的候选状态与选择指纹;可指定最多四个节点读取候选缩略图进行视觉检查,未指定则只返回状态。 Read canvas_get_settings 读取画布版本、媒体模型范围和 ComfyUI 默认连接,以及分页公开模型和连接目录。目录不含凭据和地址;更新前必须读取版本。 Read canvas_get_workflow_run 读取当前已关联画布中的确切持久工作流状态、节点进度和剩余媒体预算。 Read canvas_inspect_images 读取受限缩略图供视觉核对。默认检查正式采用图片;检查候选或历史图片时,先从 canvas_read 的 jobHistory 取得成功任务 id,再通过 versions 指定 nodeId 与 jobId。不会采用、修 Read canvas_inspect_media 读取音视频配置、候选、运行阶段、输入连线和可运行准备状态;expectedRevision 固定审核基线,失败返回 readError。不播放、不解码,metadataOnly 不表示已观看或听取内容。 Read canvas_inspect_media_content 读取画布节点真实已采用的音视频,执行技术解码并生成有界抽样供内容检查。元数据或成功解码不等于完整观看;工具只读媒体,不修改媒体。 Read canvas_list_nodes 分页枚举当前会话已关联画布的权威节点摘要;可按节点类型过滤,expectedRevision 固定审核基线。局部失败返回 readError,不代表内容错误;不返回素材 ID、媒体 URL 或本地路径。 Read canvas_list_workflow_runs 列出当前已关联画布的持久工作流摘要,不触发恢复或远端生成。 Read canvas_prepare_image_attachment 把统一媒体生成或已登记图片绑定到现有图片卡并冻结配置、真实请求和输入版本;不会创建新卡、执行模型或自动采用。生成应先准备再执行,完成后用 canvas_attach_image_result。 Read canvas_read 按 ID 读取有限节点、正文与邻接;expectedRevision 固定审核基线。逐节点 readError 表示读取失败,missingNodeIds 表示不存在,omittedEdgeCount 表示相关连线未完整 Read canvas_review_media 记录 Agent 基于实际已收到样本作出的内容评审,只写任务证据,不修改媒体。抽样不能冒充完整观看;当前只支持 sampled,full 请求由 Host 按真实能力明确拒绝。 Read canvas_view 查询或操作当前会话前台画布:选择、聚焦、适应画布、视口、选择/拖动工具、工作台展开及大小位置。query允许计划模式,其余操作需要执行权限;没有对应前台画布时明确不可用,不改变其他会话。 Read CompactContext Compact only the current Pi Agent session after this turn finishes. Before calling, persist a durable handoff Read factory_get_batch Read one persisted evaluation or comparison batch by ID. Use run IDs from its items to inspect full model outp Read factory_get_experiment Read experiment metadata and optimization details by default. Explicit validation or held-out details mark obs Read factory_get_run Read exactly one run by runId, including the original input, output, traces, frozen review criteria, evidence, Read factory_get_scene Read one scene: its current definition (inputs, outputs, steps, capabilities, model slots, review criteria) pl Read factory_list_batches List persisted batch summaries for a scene. Returns counts and quality/comparison summaries without copying fu Read factory_list_datasets List saved test datasets with their exact cases and versions. Read-only. Dataset versions change whenever a ca Read factory_list_deliveries List local capability-package deliveries and their verification state. Read-only. Read factory_list_evaluations List recent persisted evaluations for one scene, including constraint and review evidence summaries. Missing o Read factory_list_experiments List persisted experiment summaries without sample labels. Read factory_list_runs Read the recent runs of one scene, newest first: status (execution), valid (format constraints), review (autom Read factory_list_scenes 列出当前工作区的编排场景摘要(id、name、当前版本、是否有草案、步骤数)。场景是工作区内的流程,不是另一个项目;用户要求修改或增加步骤时先调用本工具,再复用匹配场景的 sceneId。只读;完整定义使用 factor Read factory_list_stubs List the local capability stubs of this workspace: capabilityId, whether it is a placeholder, and the note. Re Read factory_list_tasks Read the reusable whole-scene task inputs previously submitted by the human, newest first. Read-only. Reuse th Read factory_list_versions List a scene\ Read factory_prepare_draft Step 1 of 2 for changing a scene. Computes exactly what would change and returns a preparedId plus a frozen sn Read factory_prepare_stub Step 1 of 2 for binding a capability stub. Returns a preparedId plus an approval snapshot; nothing is written Read mcp__automation__get_automation 读取单个 DutyDeck 定时任务详情和运行记录。定时任务自动执行中可以省略 id 来读取当前任务,用于自检和自迭代。 Read mcp__automation__list_automations 列出 DutyDeck 持久化定时任务。用于查看已有长期反复任务、判断是否需要新建任务、检查运行状态和最近失败情况。 Read mcp__automation__list_workspaces 查询可作为定时任务创建目标的工作区,仅返回 ID、名称、slug、是否当前工作区和项目根状态,不读取文件内容。跨工作区创建前先查询并用精确 ID 选择;重名时向用户确认。managed 表示托管项目;missing/no Read mcp__collaboration__answer_delegation_question 代答协作子会话的阻塞问题或审批权限请求。从 delegation 的 pendingBlockedEvents 获取 blockedEventId。 Read mcp__collaboration__continue_delegation 向已完成、已失败、已取消或已中断的协作子会话追加后续指令。子会话保留完整上下文继续执行。 Read mcp__collaboration__delegate_agent 创建一个真实可见的 DutyDeck 协作子 Agent 会话来并行处理独立子任务。可选 thinkingLevel 指定子会话首轮思考强度;模型不支持时运行时会安全归一化。返回中的 configuredThinking Read mcp__collaboration__delegate_agents 批量创建多个真实可见的 DutyDeck 协作子 Agent 会话。每个 item 可独立指定首轮 thinkingLevel;模型不支持时运行时会安全归一化。返回中的 configuredThinkingLevels/ Read mcp__collaboration__get_delegation_results 按委派 ID 读取一个或多个 DutyDeck 协作子会话的结果摘要。返回中的 thinkingLevel 表示配置/请求值,不代表模型 capability normalization 后的实际运行档位。 Read mcp__collaboration__list_available_agent_models 列出当前父会话渠道下已启用、可用于协作子 Agent 的模型。需要给 delegate_agent/delegate_agents 指定 modelId 前应先调用此工具。 Read mcp__collaboration__list_delegations 列出当前父会话创建的 DutyDeck 协作子会话及状态。返回中的 thinkingLevel 表示配置/请求值,不代表模型 capability normalization 后的实际运行档位。 Read mcp__planning__acknowledge_reminder 确认并关闭一个到期提醒,不会删除 Todo 或日程。仅在用户明确要求关闭提醒时使用。仅 Pi Agent 可用。 Read mcp__planning__get_calendar_event 按 ID 读取一个日程的完整详情。仅 Pi Agent 可用。 Read mcp__planning__get_todo 按 ID 读取一个 Todo 的完整详情。仅 Pi Agent 可用。 Read mcp__planning__list_active_reminders 列出当前已到期且未确认的常驻提醒。用于帮助用户处理提醒,不用于扫描全部历史。仅 Pi Agent 可用。 Read mcp__planning__list_calendar_events 列出 DutyDeck 日程(包含用户明确连接的系统日历投影)。nativeOrigin 表示编辑会写回系统;Agent 修改前必须先取得用户确认。仅 Pi Agent 可用。 Read mcp__planning__list_groups 列出指定范围的 Todo 或日程分组。创建或归入分组前优先调用,以复用该范围内的现有分组。仅 Pi Agent 可用。 Read mcp__planning__list_tags 列出可用于 Todo 与日程的标签。创建或归类前优先调用,以复用已有标签。仅 Pi Agent 可用。 Read mcp__planning__list_todos 列出 DutyDeck Todo(包含用户明确连接的系统提醒事项投影)。返回项的 nativeOrigin 表示编辑会写回系统;对该类项单项编辑/完成先征得用户确认,批量修改和删除必须明确确认。仅 Pi Agent 可用 Read mcp__planning__snooze_reminder 将未确认提醒推迟指定分钟数。仅 Pi Agent 可用。 Read media_discover_workflows 未指定工作流时,从显式连接、画布绑定或未绑定画布的唯一启用 ComfyUI 服务器分页读取候选,检查真实节点、媒体输入数量、输出和当前服务器兼容性。复用资源快照;UI 图仅在确定映射时转换,返回阻塞原因。candidat Read media_get_asset_file 把当前项目的可信媒体资产引用解析为已验证本地文件及技术元数据,供 ffmpeg 或分析 Skill 使用;不能传入任意路径或扩大授权目录。 Read media_get_generation 读取当前 Agent 会话持有的统一媒体任务快照,不触发新的供应商请求。 Read media_get_node_schema 按已发现的准确 classType 从本地快照读取 ComfyUI 节点端口、标量约束、模型及安全选项;首次没有快照才拉取。可在编写工作流之前调用,需要更新时先调用 media_list_resources(kind=n Read media_get_run 读取属于本次 Agent 的媒体任务进度和产物摘要,不触发远端生成。 Read media_inspect_workflow 检查已保存工作流、完整定义或原始 ComfyUI API prompt,分页返回节点、绑定、输出和兼容问题。 Read media_list_api_models 列出 Host 按当前项目和 Canvas scope 过滤后的图片、音频或视频 API 模型;不自动选择第一项。 Read media_list_assets 分页列出当前项目已授权素材的稳定引用、角色和技术元数据,不返回本地路径。 Read media_list_models 列出当前配置目录中的图片、音频和视频模型;返回稳定 modelRef 与执行能力,不返回地址或凭据。 Read media_list_profiles 兼容读取当前项目的旧媒体预设;新任务优先使用公共工作流或项目草稿。 Read media_list_resources 分页查询已连接 ComfyUI 的节点、模型、工作流或 assets。节点、模型、工作流首次拉取后持久保存,默认搜索和分页读取本地快照;需要更新目录时显式设置 refresh=true。同一快照也供画布使用。保留来源能力 Read media_list_sources 列出当前会话已授权的附件来源;sourceRef 是不透明标识,不是本地路径。 Read media_list_workflows 读取画布 ComfyUI 默认绑定、本次 effectiveConnection 与本地模板摘要。未绑定且仅一台启用服务器时可直接发现工作流,不必先手动绑定;connectionSource 说明连接来源,读取不保存偏好 Read media_match_assets 按类型、显式素材 ID 与语义角色匹配工作流输入,歧义或缺失必须由调用方处理。 Read media_prepare_generation 按稳定 modelRef 准备一次图片、音频或视频生成;只校验配置和参数,不发送供应商请求。 Read media_read_remote_workflow 读取精确 descriptor 的远端工作流。小图默认返回正文;大图自动返回节点分页,可用 section=nodes/inputs/outputs/issues 和 offset/limit 精读完整结构与错误。输入页 Read media_use_remote_workflow 直接使用已发现的服务器工作流。Host 自动解析并复用不可变执行快照,返回卡片配置所需版本;无需创建、导入或发布本地模板,不提交生成。内容变化时重新发现并核对。 Read ops_data_diagnose Read bounded sanitized diagnostics. MySQL requires database scope with an explicit database and sessions/state Read ops_data_test Test one saved MySQL/PostgreSQL/SQLite source or saved Redis source using the connection retained by DutyDeck. Read ops_database_describe Read columns and indexes for one saved MySQL/PostgreSQL/SQLite table unless it is disabled in Server Ops. Post Read ops_database_query Execute one read-only MySQL, PostgreSQL or SQLite SELECT on a saved source. Every referenced base table must b Read ops_database_rows Read one bounded page from a saved MySQL/PostgreSQL/SQLite table unless disabled in Server Ops. PostgreSQL req Read ops_database_tables List tables in a saved MySQL/PostgreSQL/SQLite source, excluding disabled tables and system databases or schem Read ops_redis_read 读取已保存 Redis 连接的单个键,无需额外板块授权。支持 GET、TYPE、TTL、PTTL、EXISTS、STRLEN、HLEN、LLEN、SCARD、ZCARD、HGET、HEXISTS、SISMEMBER、ZS Read ops_resources List saved MySQL/PostgreSQL/SQLite read-only sources plus already connected saved servers and saved Redis sour Read ops_server_logs Read one bounded log snapshot from an already connected saved server. Never follows logs or starts a persisten Read ops_server_overview Read a bounded structured overview from one already connected saved server.${UNTRUSTED_EVIDENCE} Read ops_server_services Read the bounded systemd service list from one already connected saved server.${UNTRUSTED_EVIDENCE} Read proma_knowledge_asset Copy one scanned metadata-only binary asset into the approved outline item assets path. Returns a relative Mar Read proma_knowledge_maintenance_next Return one small prioritized batch of unprocessed project source references without loading their bodies. Read Read proma_knowledge_read Read a registered entry using sourceRef (preferred) or legacy entryId/expectedRevision, never both. Returns pa Read proma_knowledge_refresh_sources Refresh one to four already registered and authorized sources after a file change. It does not scan directorie Read proma_knowledge_scan_sources 在用户要求分析项目或发现新增资料时扫描当前项目并等待完成。只更新来源索引,不返回正文,也不计为 Agent 已读证据,不启动新 Agent;随后用 search/read 核对证据。不要在普通检索或每轮对话中重复扫描。 Read proma_knowledge_search Search distilled project knowledge by default. Set scope to sources to find original documents or memory files Read proma_knowledge_status Read the latest proposal awaiting user confirmation, the approved plan, pause state, and outline item IDs/revi Read proma_memory_record append a concise, user-approved Markdown memory entry to the current workspace memory directory. Only relative Read proma_workspace_list_mcp_servers List the current workspace MCP servers and their safe connection status. No credentials, headers, environment Read prototype_adopt_page 自动采纳指定页面候选并创建新的页面版本。必须先读取该页面取得pageRevision与baseRevision;只允许目标页变化,其他页面即使已有新版本也不会被覆盖,页面基线变化时拒绝采纳。 Read prototype_compare_page 比较同一页面的两个历史版本或候选,只返回该页默认HTML、状态、素材和流程差异;先读取页面和版本再决定是否采纳。 Read prototype_page_versions 读取指定页面的独立版本摘要;全局revision仅用于追溯,不会加载邻页HTML。创建新页面前先使用prototype_list与prototype_read_page判断是否已有可复用页面。 Read prototype_propose_page 局部修改优先使用。仅传入指定页HTML;stateId指定已有附加状态,省略时仅修改默认HTML并保留states;新增删除状态使用完整propose。宿主读取baseRevision并保持其他页面、目录、来源和流程原样 Read prototype_read_page 局部修改优先使用。读取指定版本的目标页面HTML与全部states、原型元信息、页面目录摘要和相关流程;不返回其他页面正文。pageId来自界面引用或已有目录,revision来自prototype_list或界面引用。 Read server_docker_action 对指定的完整容器 ID 启动、停止或重启容器;每次都需要单独确认。执行前会再次核对容器身份并检查结果,结果不明确时不会自动重试。 Read server_docker_detail 根据完整的 64 位容器 ID 查看已脱敏的容器详情;不会返回环境变量和秘密。 Read server_docker_resources 查看已连接服务器本机 Docker 守护进程中的容器、镜像、网络和卷;结果有数量和大小限制。 Read server_exec 在已连接的服务器上执行一条受范围限制的非交互式命令(不会等待键盘输入)。只读诊断可能自动执行;会修改服务、文件或数据的命令,以及无法判断风险的命令,每次都要单独确认。 Read server_files_list 查看已连接服务器上的一页远程目录。结果最多 64 KiB;如果被截断,就不代表完整目录列表。 Read server_files_mutate 对远程文件执行一次精确变更;每次都需要单独确认。覆盖已有文件必须提供 server_files_read 返回且文件未变化的 editToken;不允许递归删除、修改权限或自动重试。 Read server_files_read 仅在用户要求时读取远程文件的 UTF-8 文本预览。不会跟随符号链接,也不会把二进制文件或超大内容作为文本返回;结果最多 64 KiB。 Read server_list 列出本次 Agent 运行开始时已连接的已保存服务器,返回公开 SSH 身份和连接阶段。 Read server_status 查看已连接的已保存服务器的公开 SSH 连接状态。 Read server_task_interrupt 逐次确认后关闭当前会话指定任务的独立 SSH channel。若关闭无法确认,会断开对应 SSH 连接以清理失管通道。远程子进程可能继续运行,此操作不承诺回滚,必须核对最终状态。 Read server_task_read 读取当前会话任务的有界输出和真实退出状态。首次 offset=0,后续使用 nextOffset;waitMs 最多 30000,等待新输出或终态而无需 Bash sleep。exited 且 exitCode=0 才表 Read TaskGet 读取某个可见进度任务的当前状态。 Read TaskList 列出当前 turn 中已创建的可见进度任务。 Read TerminalList List terminals owned by the current Agent session, including cwd and running/exited state. Use this before eve Read TerminalRead Read bounded buffered output from one current-session Agent-owned terminal. By default returns the latest 12,0 Read TodoRead 读取当前 turn 的任务列表。兼容 Claude SDK 的 TodoRead。 Read VisionRelay Use this when the current DeepSeek V4 model needs to understand an uploaded or authorized image. It sends one Read visual_production_delegate_review 在当前已打开的视觉范围中启动独立受限审图子 Agent;只能查看授权素材并保存观察,不会生成媒体或继承主会话图片。返回回执后用read_review查询。 Read visual_production_read_review 读取本轮专用检查子任务真实终态。completed才返回固定范围的持久观察;摘要不等于验收证明。旧运行只可从重新打开范围的历史观察追溯,不能自动续跑。 Read WebFetch Fetch and extract readable Markdown content from a URL through DutyDeck\ Read WebSearch Search the web for up-to-date information through DutyDeck\

Related servers

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

Questions about DutyDeck

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

Yes. The DutyDeck server exposes 10 destructive tools including canvas_cancel_media_run, factory_cancel_experiment, mcp__automation__delete_automation. 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 DutyDeck? +

The DutyDeck server has 80 write tools including api_bind_crypto_profile, api_declare_variables, api_save_crypto_profile. 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 DutyDeck.

How many tools does the DutyDeck MCP server expose? +

247 tools across 4 categories: Destructive, Execute, Read, Write. 129 are read-only. 118 can modify, create, or delete data.

How do I enforce a policy on DutyDeck? +

Register the DutyDeck 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 DutyDeck tool call.

Deterministic rules across all 247 DutyDeck tools. Per-identity grants. Full audit log. Live in minutes. Nothing to install.

Instant setup, no code required.

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

// WHERE THIS COMES FROM

These policies come from DutyDeck'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 →

// GET IN TOUCH

Have a question or want to learn more? Send us a message.

Message sent.

We'll get back to you soon.