TRAECNclaw MCP
TRAECNclaw MCP Skill
정식 TRAECNclaw 0.6.0에서 커밋
40bcf7707ed9b0ff7938aea8e511d0b87d300705기준으로 생성됨.
TRAECNclaw는 사용자의 Mac에서 TraeCN 데스크톱 앱을 운영하기 위한 로컬 우선(local-first) MCP 서버이자 Agent Skill입니다. 이 저장소는 공개 배포 미러로, 휴대용 Skill, 설치 가능한 서버 아카이브, 출처(provenance), 마켓플레이스 메타데이터를 제공하며 비공개 정식 이력은 게시하지 않습니다.
TRAECNclaw는 MCP 계약 버전 5를 노출하며 정확히 20개의 안정적인 단일 의도(single-intent) 도구를 제공합니다. 일반적인 Agent 흐름은 다음과 같습니다:
필요할 때만 워크스페이스를 엽니다.
필요할 때만 모델 또는 모드를 선택합니다.
메시지 하나를 보내고 호출자 컨텍스트를 해제합니다.
레거시 MCP 클라이언트는 지속 알림(durable notification)을 받고, 최신
2026-07-28호스트는 Tasks를 협상하고 반환된 정확한 작업 ID를 구독할 수 있습니다.
대기열, 대기, 복구, 일상적인 비명령 질문, 작업 소유권, 지속 완료 처리는 게이트웨이에서 관리됩니다.
로컬 서버는 개행 구분 stdio를 사용하고 MCP 2026-07-28을 구현하며, 초기화 기반 2025-11-25 및 2024-11-05 클라이언트와도 호환됩니다. TRAECNclaw 계약 버전 5는 도구 표면 버전이지 MCP 프로토콜 개정판이 아닙니다. 독립적으로 협상되는 상위 초안(upstream-draft) io.modelcontextprotocol/tasks 확장은 도구를 추가하지 않습니다.
서버 설치
검증된 한 가지 채널을 통해 정확한 릴리스를 설치하세요:
npm install --global @luckycat133/traecnclaw@0.6.0
# or, after `brew info` reports 0.6.0
brew install Luckycat133/tap/traecnclaw해당 GitHub Release에서도 전체 스코프 npm tarball과 결정적 MCPB 번들을 제공합니다. 무관한 비스코프 traecnclaw@0.3.1 패키지는 과거 이력이며 현재 설치 경로가 아닙니다.
Related MCP server: nuphus-mcp
Agent Skill 설치
공개 GitHub 저장소에서 오픈 Skills CLI로 설치하세요:
npx skills add https://github.com/Luckycat133/traecnclaw-mcp-skill \
--skill traecnclaw-mcp \
-g또는 GitHub Release에서 해당 Skill 아카이브를 다운로드하세요. Skill은 .codex/skills/traecnclaw-mcp에 있으며 assets/mcp-client-config.json, scripts/setup-mcp.js, 서버 실행기를 포함합니다. Skill 설치 또는 업데이트 후 Agent 호스트를 재시작하세요.
scripts/setup-mcp.js를 사용하여 서버 검색을 검증하고 호스트 항목을 생성하세요. 일반적인 런타임은 다음과 같습니다:
Agent host -> local stdio server -> gateway on the same Mac -> TraeCN서버와 게이트웨이는 TraeCN과 동일한 사용자 소유 Mac에서 실행됩니다. 마켓플레이스 컨테이너는 stdio 스키마를 검사할 수 있지만, 클라우드 컨테이너가 사용자의 로컬 TraeCN 데스크톱을 투명하게 제어할 수는 없습니다.
준비된 공식 레지스트리 메타데이터는 로컬 stdio 전송과 함께 스코프 패키지 @luckycat133/traecnclaw@0.6.0을 사용합니다. Smithery Local은 일치하는 검증된 MCPB 번들을 사용하며, 폐기된 smithery.yaml 형식은 계속 제공되지 않습니다.
구성
게이트웨이 기본값은 TRAECN_GATEWAY_HOST=127.0.0.1 및 TRAECN_GATEWAY_PORT=8788입니다. 루프백이 아닌 바인딩에는 TRAECN_GATEWAY_TOKEN이 필요합니다. MCP 도구 프로필 설정은 없습니다.
검색 채널
동일한 공개 저장소가 Skill 및 MCP 디렉터리 제출의 정식 소스입니다. 디렉터리 목록은 해당 버전, 로컬 stdio 전송, macOS 요구 사항, 20개 도구 수, 생성된 설치 명령이 이 Release와 일치할 때만 신뢰할 수 있습니다.
선정된 채널에는 ClawHub, Glama, skills.sh, AwesomeSkills.dev, MCP.Directory, MCPB/Smithery Local이 포함되며, 현재 공개 패키지가 존재한 후에는 공식 MCP 레지스트리와 PulseMCP도 포함됩니다. 다른 디렉터리는 포크를 유지하기보다 동일한 출처와 설치 사본을 재사용해야 합니다.
Glama
glama.json과 생성된 루트 Dockerfile은 디렉터리 등록, 관리자 검증, 보안/품질 스캔, 도구 스키마 검사를 지원합니다. Dockerfile은 현재 공개 Release를 설치하며 게이트웨이 토큰이나 모의 브리지를 절대 포함하지 않습니다.
TRAECNclaw는 로컬 stdio 서버로 등록되어야 합니다. Glama 호스팅 컨테이너를 사용자 Mac에 대한 자동 원격 연결로 설명하지 마세요.
출처 및 채널 준비 상태는 SOURCE_REVISION 및 release-manifest.json을 참조하세요.
Available Tools
20 toolstraecn_answer_questionAnswer TraeCN QuestionA
Resolve an exceptional question from TraeCN that the gateway could not handle mechanically by supplying the user's decision. The gateway routes the answer to the pending visible prompt and reports the outcome. Only call when a pending interaction exists.
Inputs: answer (required). Returns: Interaction resolution outcome with optionally resumed tasks.
traecn_cancel_taskCancel TraeCN TaskADestructive
Cancel one gateway-tracked task by ID. The gateway stops shepherding the work at its next control point and records the cancellation; results already produced remain in task history. Returns the task identity and post-cancellation status.
Inputs: taskId (required). Returns: Cancelled task identity and any visible-generation stop outcome.
traecn_create_conversationCreate TraeCN ConversationA
Prepare a fresh TraeCN conversation for a genuinely new context. Returns a native conversation identity when TraeCN has allocated it; an empty draft returns conversationId=null and creationDeferred=true until its first message. Reuses an already-empty draft; otherwise creates another context. Call once per new context.
Inputs: no named parameters. Returns: Fresh conversation or empty draft awaiting a native identity.
traecn_decide_approvalDecide TraeCN ApprovalADestructive
Approve or deny the unsafe or ambiguous command card currently visible in TraeCN. Approval revalidates the exact expectedCommand against the card and requires acknowledgeRisk plus an audit reason recorded locally; denial needs only the decision. Never approve beyond the scope the user authorized.
Inputs: decision (required), reason, acknowledgeRisk, expectedCommand. Returns: Approval decision outcome with optionally resumed tasks.
traecn_delete_conversationDelete TraeCN ConversationADestructive
Permanently delete one inactive TraeCN conversation by stable identifier. Requires the exact expectedTitle returned by traecn_list_conversations plus an explicit irreversible-deletion acknowledgement; there is no gateway recovery path.
Inputs: expectedTitle (required), conversationId (required), acknowledgePermanentDeletion (required). Returns: Deletion outcome, or title/active-state mismatch details.
traecn_get_taskGet TraeCN TaskARead-onlyIdempotent
Read one known gateway task by ID: status plus final result, or the full execution trace with detailLevel 'trace' for diagnosis. Call once when notified of completion or once later when resuming after a disconnect. Not a progress primitive — do not poll in a tight loop.
Inputs: taskId (required), detailLevel. Returns: Task status snapshot with final result or diagnostics.
traecn_list_conversationsList TraeCN ConversationsARead-onlyIdempotent
List the conversations available in the current TraeCN mode with stable identifiers accepted by conversation-scoped tools. Read-only; call when choosing or verifying conversation context.
Inputs: no named parameters. Returns: Conversations for the current mode with stable identifiers.
traecn_list_modelsList TraeCN ModelsARead-onlyIdempotent
List the model names currently offered in the TraeCN UI and the current selection. If Auto hides manual models, report manualModelsHidden without changing modes. Call before select_model whenever the model choice matters.
Inputs: no named parameters. Returns: Current model plus available model names.
traecn_list_setting_optionsList TraeCN Setting OptionsARead-onlyIdempotent
List the visible choices of one dropdown-style setting identified by section and label. Required before select_setting_option; never invent an option value.
Inputs: label (required), section (required). Returns: Selectable choices for one dropdown setting.
traecn_list_settingsList TraeCN Section SettingsARead-onlyIdempotent
List the individual settings and their control types within one section returned by traecn_list_setting_sections. Read-only discovery preceding any setting mutation; never guess a label.
Inputs: section (required). Returns: Settings and control types within one section.
traecn_list_setting_sectionsList TraeCN Settings SectionsARead-onlyIdempotent
List the settings sections currently visible in TraeCN's settings UI. Required first step for any settings operation; never guess a section name.
Inputs: no named parameters. Returns: Visible settings section names.
traecn_open_workspaceOpen TraeCN WorkspaceA
Open one absolute local folder path as the active TraeCN workspace and verify it became active before returning. Call before submitting work that depends on a specific project; afterwards do not repeat the path inside the task message.
Inputs: path (required). Returns: Opened workspace identity with launch and verification details.
traecn_select_conversationSelect TraeCN ConversationA
Make one existing conversation the active target for subsequent visible interaction. Returns the confirmed active conversation identity.
Inputs: conversationId (required). Returns: Confirmed active conversation identity.
traecn_select_modeSelect TraeCN ModeAIdempotent
Switch TraeCN between Solo and IDE mode and confirm the resulting active mode. Switching to the current mode succeeds as a no-op. Call only when the requested mode differs.
Inputs: mode (required). Returns: Active mode after the switch.
traecn_select_modelSelect TraeCN ModelAIdempotent
Select one model by exact name from traecn_list_models and confirm the picker state before returning. Selecting the already-active model succeeds as a no-op. Does not wait for model availability or capacity.
Inputs: model (required). Returns: Applied selection with before/after models and queue readiness.
traecn_select_setting_optionSelect TraeCN Setting OptionA
Select one previously listed option for a dropdown setting identified by section and label, and confirm the applied selection. Only values returned by traecn_list_setting_options are valid.
Inputs: label (required), value (required), section (required). Returns: Applied dropdown selection and driver outcome.
traecn_send_messageSend TraeCN TaskA
Submit one user instruction to TraeCN as durable background work and return its stable taskId immediately; the gateway owns queueing, retries, recovery, and result persistence. Use this for every delegated coding request, optionally targeting an existing conversation via conversationId. Do not send follow-up messages to probe progress — track the returned task instead.
Inputs: message (required), conversationId. Returns: Accepted task identity, original terminal replay, or gateway rejection with rea…
traecn_set_setting_textSet TraeCN Setting TextA
Enter one text value into the setting field identified by section and label, and confirm what the field contains afterwards.
Inputs: text (required), label (required), section (required). Returns: Applied text value and driver outcome.
traecn_set_setting_toggleSet TraeCN Setting ToggleAIdempotent
Enable or disable one toggle setting identified by section and label, then read back the applied state. Applying the already-current state succeeds as a no-op.
Inputs: label (required), enabled (required), section (required). Returns: Applied toggle value and driver outcome.
traecn_stop_generationStop Visible GenerationADestructive
Stop only the generation currently visible in the named active TraeCN conversation. Visible output may belong to the user or another agent, so this requires acknowledgeUntrackedWork=true plus an audit reason. Prefer cancel_task whenever a gateway task ID exists.
Inputs: reason (required), conversationId (required), acknowledgeUntrackedWork (required). Returns: Verified scoped stop outcome, or the active-conversation mismatch details.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 20 tools
The tool set is largely cleanly partitioned: tasks, conversations, settings, models/modes, workspace, and interaction handling each have their own verbs. The only genuinely fuzzy pair is cancel_task vs stop_generation, but the description explicitly tells the agent to prefer cancel_task when a task ID exists, and the settings trio (toggle/option/text) is distinguished by control type. A few tools (list_settings vs list_setting_options vs list_setting_sections) are close cousins but the descriptions establish a clear required ordering.
Every tool uses the uniform traecn_<verb>_<noun> snake_case pattern (list_, select_, set_, create_, delete_, get_, cancel_, stop_, send_, open_, answer_, decide_). Verbs are used semantically consistently: list_ for read-only discovery, select_ for choosing from a listed set, set_ for applying a value. No camelCase or style mixing anywhere.
Twenty tools is on the heavy side, but the server spans several distinct sub-domains (task lifecycle, conversation lifecycle, settings discovery/mutation, model/mode selection, approvals, workspace), so most tools earn their place. It is borderline rather than bloated, and consolidation of the four settings tools would make it tighter.
Coverage is strong across most lifecycles: tasks (send/get/cancel), conversations (list/select/create/delete), settings (sections/list/options/set/select), models and mode, workspace opening, and interactive prompts (answer_question, decide_approval). Minor gaps remain — there is no list_tasks to enumerate tracked work, no conversation rename, and no workspace listing/current-workspace read — but agents can work around these using the returned taskId.
Maintenance
Related MCP Connectors
One MCP server for all your MCP servers: code mode, per-tool permissions, approvals and audit.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
Hosted remote MCP for protected account access, permissions, activity history and supported undo.
Headless task management for bots, apps and AI agents. Remote MCP server, OAuth or access token.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceEnables automation of native Windows desktop applications through screen capture, mouse/keyboard control, and waiting for UI changes, exposing them as MCP tools.1-
- AlicenseNot gradedqualityCmaintenanceProvides desktop automation capabilities (e.g., click, browser snapshot) as MCP tools for DSH agents, enabling computer use via natural language.592 npm5MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to perform deterministic Windows desktop and browser automation through MCP, using pre-validated UI Automation and DOM locators for fast, stable execution of ERP and business workflows.1MIT
- AlicenseNot gradedqualityAmaintenanceEnables MCP clients to automate live browser sessions by having an LLM plan one outcome at a time while a hosted decision model picks elements, actions, values, and completion status. It exposes tools for navigation, direct actions, page reading/snapshots, checks/choices, screenshots, and calling WebMCP tools registered by pages.1,464 npmMIT