TRAECNclaw

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:

1. Open a workspace only when needed.
2. Select model or mode only when needed.
3. Send one message and release the caller context.
4. Legacy MCP clients receive a durable notification; modern 2026-07-28 hosts 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:

npm install --global @luckycat133/[email protected]
# or, after `brew info` reports 0.6.0
brew install Luckycat133/tap/traecnclaw

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.

Install the Agent Skill

Install from the public GitHub repository with the open Skills CLI:

npx skills add https://github.com/Luckycat133/traecnclaw-mcp-skill \
  --skill traecnclaw-mcp \
  -g

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:

Agent host -> local stdio server -> gateway on the same Mac -> TraeCN

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.