GLM Claude Code Worker MCP

Coding-worker MCP for Codex Desktop / Claude Code. Codex plans, delegates, and reviews; GLM-5.1 does the expensive code reading, editing, and checking through Claude Code.

Languages: English | 简体中文

What It Does

Runs real coding work with GLM-5.1 through Claude Code CLI
Works separately from your normal Claude Code usage
Async worker tools: start / get / tail / wait / cancel
Returns status, changed files, snapshot diffs, policy, and checks
Uses scoped Claude Code permissions by default
Includes setup / doctor for install, auth, and environment checks

Quick Start

git clone https://github.com/YOUR_USERNAME/glm-claude-code-worker-mcp.git
cd glm-claude-code-worker-mcp
npm install
node src/glm-worker-mcp.mjs --setup
node src/glm-worker-mcp.mjs --doctor

Or use without cloning:

{
  "mcpServers": {
    "glm-code-worker": {
      "command": "npx",
      "args": ["github:YOUR_USERNAME/glm-claude-code-worker-mcp"]
    }
  }
}

Source-mode MCP config (recommended for local development):

{
  "mcpServers": {
    "glm-code-worker": {
      "command": "node",
      "args": ["/absolute/path/to/glm-claude-code-worker-mcp/src/glm-worker-mcp.mjs"]
    }
  }
}

Requirements

Node.js 20+
Claude Code CLI (npm install -g @anthropic-ai/claude-code)
GLM API key from open.bigmodel.cn
macOS / Linux; Windows is best-effort

Auth

Three ways to provide your GLM API key:

Environment variable: set ANTHROPIC_AUTH_TOKEN
Key file: set GLM_API_KEY_FILE pointing to a file with the key
Default file: save to ~/.codex/secrets/glm_api_key

Run node src/glm-worker-mcp.mjs --setup to configure interactively.

Tools

| Tool | Mode | Description | |---|---|---| | glm_implement_in_workspace | Sync | Synchronous worker for tiny edits | | glm_start_implementation | Async | Start a background coding job | | glm_get_job | Query | Read compact job status | | glm_tail_job | Query | Read running-job progress | | glm_wait_for_job | Query | Observe until completion or timeout | | glm_cancel_job | Action | Cancel a running job |

Use Cases

| use_case | Model | Thinking | Best for | |---|---|---|---| | auto | GLM-4.5-air | enabled | general implementation | | fast_patch | GLM-4.5-air | disabled | small patches | | simple_agent_task | GLM-5.1 | enabled | straightforward agentic coding | | scaffold_or_tests | GLM-4.5-air | enabled | scaffolding, tests, glue code | | debug_loop | GLM-5.1 | enabled | reproduce, fix, validate | | agentic_coding | GLM-5.1 | enabled | multi-step implementation | | complex_reasoning | GLM-5.1 | enabled | architecture, hard logic | | long_context_codebase | GLM-5.1 | enabled | broad codebase work | | docs_generation | GLM-5.1 | enabled | documentation |

Explicit model, thinking, or reasoning_effort values override the preset.

Verification

node src/glm-worker-mcp.mjs --doctor   # Check environment
node scripts/smoke-stream-events.mjs    # Verify stream parsing
node scripts/smoke-tools-list.mjs       # Verify tool list

Architecture

bin/claude-glm.mjs         Claude Code bridge launcher
  -> sets ANTHROPIC_BASE_URL=https://open.bigmodel.cn/api/anthropic
  -> sets ANTHROPIC_MODEL=GLM-5.1
  -> spawns Claude Code CLI

src/glm-worker-mcp.mjs     MCP server (tools, job lifecycle, diffs)
src/core/config.mjs        Constants, USE_CASES, WORKER_PROFILES
src/core/stream-events.mjs Claude Code stream-json event parsing

License

MIT

MCP Servers