An MCP (Model Context Protocol) server intended to run as a Warp or other CLI Based AI Assistant plugin/tool.
It provides a thin layer for planning/orchestration by calling z.ai using your Devpack coding plan API key, so more work can happen inside the MCP process (minimizing Warp usage / context window).
- Node.js 18+ (for built-in
fetch) - Git
- Copy
.env.exampleto.envand fill in your key. - Install deps:
npm install
- Build:
npm run build - Run:
npm start
- Build the server so
dist/exists:npm run build
- In Warp, open Settings → AI → Manage MCP servers (or Warp Drive → Personal → MCP Servers) and click Add.
- Paste a JSON config like this (update
working_directoryto your checkout path):
{
"mcpServers": {
"warpos": {
"command": "npm",
"args": ["start"],
"working_directory": "C:\\Development\\warpos",
"env": {},
"start_on_launch": true
}
}
}Notes:
- This server loads
.envviadotenv, so you can usually leaveenvempty as long as.envis present inworking_directory. - If you prefer not to use
.env, setDEVPACK_CODING_PLAN_API_KEY(orZAI_API_KEY) in theenvobject instead.
zai_plan: sends a prompt to z.ai and returns the response text.
This repository is being extended to support a deterministic workflow:
- Warp authors an MCD (Main Context Document)
- MCP applies Domain Profiles (behavioral/technical/methodological frameworks)
- MCP uses Atomic Task Templates (stable library) to generate project-specific Task Instances
- MCP executes work via a two-phase model: prepare (write plan/spec for inspection) → execute (apply after GO/NO-GO)
Docs:
- Domain Profiles:
docs/domain-profiles.md - MCD:
docs/mcd.md - Atomic Task Templates:
docs/task-templates.md - Task Instances:
docs/task-instances.md
- The z.ai HTTP endpoint/shape may differ from the placeholder implementation in
src/zaiClient.ts. AdjustZAI_BASE_URLand request/response mapping as needed.