TZ

tizerluo/zcode-open-bridge

Developer tools

This page was translated automatically. Technical details may differ from the original. View the original

21 stars 1 forks Quality 75 Trend 24MIT

An unofficial community project integrating ZCode (Zhipu Z.AI Agentic CLI) into the Open Agent Ecosystem (MCP / ACP)

Overview

An unofficial community project integrating ZCode (Zhipu Z.AI's Agentic Coding CLI) into the Open Agent ecosystem (MCP / ACP). ZCode is an AI programming agent produced by Zhipu Z.AI, powered by the GLM series models. It is very powerful, but by default it is a "closed-source island"—it can only be used in its own desktop app and cannot be standardized by other agents, editors, or scripts. - ZCode installed (must include CLI, included in the app) - Node.js ≥ 18 (zcode.cjs shebang is #!/usr/bin/env node, CLI executed by node) - Python 3.8+ (standard library only, zero third-party dependencies) - Login via ZCode (credentials exist ~/.zcode/v2/config.json) And in the shell configuration (~/.zshrc), configure credentials (dynamic read, unspecified key): ⚠️ (Overall review security discovery 1): eval + unescaped interpolation is a dangerous template—if config.json contains metacharacters like ", $, etc., eval will execute it, so the example above uses shlex.quote for every value. A safer approach is **: directly use the project's shared/credentials.py (three components are built-in, pure json.load reads config, without going through eval/shell), or refer to --print-injected-env' diagnostic output for manual export. - Single Q&A: --prompt " " - long-distance task (setting session goal): --target " ", or equivalently starting the prompt with /goal:--prompt "/goal" (both mutually exclusive) - Permissions: --mode plan (read-only) / build / edit / yolo (fully automatic) - Output: plain text or --json (including sessionId/response/usage) :Review systems no longer use the plan mode—plan only prohibits "file modification," but reads and explores and lets sub-agents proceed (mainly due to rate limit timeouts). Moreover, the planning inertia of plan→build tends to turn review into "reviewing and repairing simultaneously."

Overview (machine translated)

Unofficial community project integrating ZCode (Zhipu Z.AI’s Agentic Coding CLI) into the open Agent ecosystem (MCP/ACP).

What is this?

ZCode is an AI programming agent produced by Zhipu Z.AI, powered by the GLM series models. It is very powerful, but by default it is a “closed-source island”—it can only be used in its own desktop app and cannot be standardized by other agents, editors, or scripts.

This project integrates ZCode into an open ecosystem through three lightweight components:

Three components

| Component | Function | Maturity | Document | |------|------|:------:|------| | zcode-agent-help | Capability Discovery Manual: Understand All ZCode Capabilities in One Call | ✅ stable | packages/agent-help | | zcode-mcp-server | Exposing ZCode as a standard MCP server for calls by MCP clients (Claude Code/Cursor/itself) | ✅ stable | packages/mcp-server | | zcode-acp-bridge | Bridge ZCode to ACP Agent for editors like Zed/JetBrains to call | ⚠️ experimental | packages/acp-bridge | | zcode-review-gate | PR automatic review gate: poll open PR, new head tunes zcode_pr_review to review and repost verdict comments | ⚠️ experimental | packages/review-gate |

Get Started Quickly

Prerequisites

  • ZCode installed (must include CLI, included in the app)
  • Node.js ≥ 18 (The shebang of zcode.cjs is #!/usr/bin/env node, and the CLI is executed by node)
  • Python 3.8+ (standard library only, zero third-party dependencies)
  • Logged in via ZCode (credential exists ~/.zcode/v2/config.json)

Make the zcode command available in the terminal

The ZCode CLI is hidden inside the app and is not by default in the PATH:

And configure credentials (dynamic read, unclear key) in shell configuration (~/.zshrc):

⚠️ > Safety Description (Overall review of Security Discoveries 1): eval + unescaped interpolation is a hazard template—if the value of config.json contains ", ’ ‘、$等元字符会被eval执行,所以上面示例对每个值都做了shlex.quote’.

This page was translated automatically. Technical details may differ from the original.

Install

This server does not publish a one-line install command.

Open the repository installation guide New to MCP? Start with the mcpskill.org guide →