glm-mcp

glm-mcp

GLM (Zhipu / Z.ai) as a cheap delegate for any AI coding agent, exposed as an MCP (Model Context Protocol) server over stdio. This is the standalone server behind glm-mcp-claude and glm-mcp-copilot — usable by any MCP client (Claude Desktop, Cursor, Windsurf, Glama, …).

Quickstart (any MCP client)

{
  "mcpServers": {
    "glm": {
      "command": "npx",
      "args": ["-y", "glm-mcp"],
      "env": { "GLM_API_KEY": "YOUR_ZAI_KEY" }
    }
  }
}

Tools

• glm_agent — run GLM as a file-accessing coding agent (read/write/edit/run) on GLM tokens.
• glm_delegate — text-in/text-out subtask on GLM (no file access).
• glm_recommend — free advisory: GLM vs Opus, which model, and why.
• glm_status — free status: peak window, model picks, usage ledger, config health.

Config

The server boots and answers MCP introspection without an API key, but every GLM call needs GLM_API_KEY (your Z.ai / Zhipu GLM Coding Plan key — https://z.ai). See .env.example for all tuning knobs (model picks, peak window, token cap, cost bias, …).

Full docs

Repo + installers + design notes: https://github.com/djerok/glm-mcp

License

MIT © djerok