
Traecnclaw
io.github.Luckycat133由 GitHub 帳號 Luckycat133 發布v0.6.0更新於 Oct 24, 2026
Local persistent TraeCN coding-task gateway with a focused MCP Agent surface.
概覽
一個本機 MCP 伺服器,讓助理透過本機閘道操作 Mac 上的 TraeCN 桌面程式設計應用程式。
- 可以做什麼
- 視需要開啟 TraeCN 工作區,並為工作階段選擇模型或模式。向 TraeCN 送出單一訊息後釋放呼叫端情境,由閘道負責排隊、等待與復原。接收持久化的完成通知,或在較新的主機上協商 Tasks 並訂閱回傳的工作 ID。透過本機 stdio 提供固定的 20 個單一意圖工具。
- 適用情境
- 當你希望 AI 助理把程式設計工作交給同一台 Mac 上執行的 TraeCN 桌面應用程式,並在工作完成時收到通知時使用。它不適合無法連到使用者本機桌面的雲端或容器主機。
- 執行需求
- 以本機程序在 macOS 上與 TraeCN 桌面應用程式一起執行,透過 npm 套件或 Homebrew tap 安裝。閘道執行在同一台 Mac 上,預設 TRAECN_GATEWAY_HOST 為 127.0.0.1、TRAECN_GATEWAY_PORT 為 8788。綁定非回送位址需要 TRAECN_GATEWAY_TOKEN。安裝或更新後需重新啟動代理主機。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Traecnclaw,將其新增到工作區。
- 為需要使用其工具的對話啟用該服務。
Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。
其他 MCP 客戶端
參照 儲存庫 中的啟動說明。
README
TRAECNclaw MCP Skill
Generated from canonical TRAECNclaw 0.6.0 at commit
40bcf7707ed9b0ff7938aea8e511d0b87d300705.
TRAECNclaw is a local-first MCP server and Agent Skill for operating the TraeCN desktop app on the user's Mac. This repository is the public distribution mirror: it carries the portable Skill, installable server archives, provenance, and marketplace metadata without publishing the private canonical history.
TRAECNclaw exposes MCP contract version 5 with exactly 20 stable, single-intent tools. The normal Agent flow is:
- Open a workspace only when needed.
- Select model or mode only when needed.
- Send one message and release the caller context.
- Legacy MCP clients receive a durable notification; modern
2026-07-28hosts can negotiate Tasks and subscribe to the exact returned task ID.
Queueing, waiting, recovery, routine non-command questions, task ownership, and durable completion remain gateway-managed.
The local server uses newline-delimited stdio, implements MCP 2026-07-28, and
remains compatible with initialization-based 2025-11-25 and 2024-11-05
clients. TRAECNclaw contract version 5 is the tool-surface version, not the MCP
protocol revision. The independently negotiated, upstream-draft
io.modelcontextprotocol/tasks extension does not add tools.
Install the server
Install the exact release through one verified channel:
The matching GitHub Release
also provides the complete scoped npm tarball and deterministic MCPB bundle.
The unrelated unscoped [email protected] package is historical and is not a
current install path.
Homebrew compatibility transition
This repository also carries Formula/traecnclaw.rb, copied byte-for-byte from
the existing public 0.6.0 tap. Luckycat133/homebrew-tap remains available during
the transition. Existing installations do not change automatically. See
migration, checks and rollback before changing a tap
remote. The Skill, source revision and historical release downloads are unchanged.
Install the Agent Skill
Install from the public GitHub repository with the open Skills CLI:
Or download the matching Skill archive from the GitHub Release. The Skill lives
at .codex/skills/traecnclaw-mcp and includes
assets/mcp-client-config.json, scripts/setup-mcp.js, and the server launcher.
Restart the Agent host after installing or updating the Skill.
Use scripts/setup-mcp.js to validate server discovery and generate the host
entry. The normal runtime is:
The server and gateway run on the same user-owned Mac as TraeCN. Marketplace containers may inspect the stdio schema, but a cloud container cannot transparently control the user's local TraeCN desktop.
Prepared Official Registry metadata uses the scoped package
@luckycat133/[email protected] with local stdio transport. Smithery Local
uses the matching verified MCPB bundle; the retired smithery.yaml format
remains absent.
Configuration
The gateway defaults are TRAECN_GATEWAY_HOST=127.0.0.1 and
TRAECN_GATEWAY_PORT=8788. Non-loopback binds require
TRAECN_GATEWAY_TOKEN. There is no MCP tool-profile setting.
Discovery channels
The same public repository is the canonical source for Skill and MCP directory submissions. A directory listing is trustworthy only when its version, local stdio transport, macOS requirement, 20-tool count, and generated install command match this Release.
Selected channels include ClawHub, Glama, skills.sh, AwesomeSkills.dev, MCP.Directory, MCPB/Smithery Local, and—after a current public package exists—the Official MCP Registry and PulseMCP. Other directories should reuse the same provenance and install copy rather than maintaining forks.
Glama
glama.json and the generated root Dockerfile support directory registration,
maintainer verification, security/quality scanning, and tool-schema inspection.
The Dockerfile installs the current public Release and never contains a gateway
token or mock bridge.
TRAECNclaw should be listed as a local stdio server. Do not describe a Glama hosted container as an automatic remote connection to the user's Mac.
See SOURCE_REVISION and release-manifest.json for provenance and channel
readiness.
來源:README.md,提交 2407f57
工具
0版本歷史
1- v0.6.0最新Sep 16, 2026
評分
已滿足適用檢查項的 50%
該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。
- 近 30 天內可連通25 分
請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。
- 已宣告認證方式10 分
請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。
- 已列出工具15 分
請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。
- 描述必需15 分
- 圖示10 分
- 文件15 分
- 授權條款10 分