@zhixuan92/multi-model-agent-mcp 0.1.2 → 0.1.4

package/README.md

@@ -2,22 +2,9 @@
 
 MCP stdio server for [`multi-model-agent`](https://github.com/zhixuan312/multi-model-agent). Exposes one tool — `delegate_tasks` — that runs work in parallel across multiple LLM providers (Claude, Codex, OpenAI-compatible) and auto-routes each task to the cheapest provider that can handle it.
 
-## Install
+## How it works
 
-```bash
-# Run without installing
-npx @zhixuan92/multi-model-agent-mcp serve
-
-# Or install globally
-npm install -g @zhixuan92/multi-model-agent-mcp
-multi-model-agent serve
-```
-
-If you plan to use `openai-compatible` providers, also install the optional peer dependencies:
-
-```bash
-npm install -g @zhixuan92/multi-model-agent-mcp @openai/agents openai
-```
+You don't run this server yourself. Your MCP client (Claude Code, Claude Desktop, Cursor, …) spawns it over stdio whenever a session starts, using the config snippet below. No install step, no long-running process to manage — `npx` fetches the latest version on demand.
 
 Requires Node `>= 22`.
 
@@ -63,24 +50,28 @@ Provider auth:
 - **`claude`** uses `ANTHROPIC_API_KEY` if set, otherwise the local Claude auth flow
 - **`openai-compatible`** uses `apiKeyEnv` (preferred) or inline `apiKey`
 
-## Register with an MCP client
+## Setup
 
 ### Claude Code
 
+One command — the client will spawn the server on demand. Use `-s user` so the server is available in **every** project on your machine, not just the directory where you ran the command:
+
 ```bash
-claude mcp add multi-model-agent -- npx @zhixuan92/multi-model-agent-mcp serve
+claude mcp add multi-model-agent -s user -- npx -y @zhixuan92/multi-model-agent-mcp serve
 ```
 
 If your providers need environment variables:
 
 ```bash
-claude mcp add multi-model-agent \
+claude mcp add multi-model-agent -s user \
   -e OPENAI_API_KEY=sk-... \
   -e ANTHROPIC_API_KEY=sk-ant-... \
   -e MINIMAX_API_KEY=... \
-  -- npx @zhixuan92/multi-model-agent-mcp serve
+  -- npx -y @zhixuan92/multi-model-agent-mcp serve
 ```
 
+Without `-s user`, `claude mcp add` defaults to local scope and only registers the server in the current project.
+
 ### Claude Desktop
 
 Add to `claude_desktop_config.json`:
@@ -90,7 +81,7 @@ Add to `claude_desktop_config.json`:
   "mcpServers": {
     "multi-model-agent": {
       "command": "npx",
-      "args": ["@zhixuan92/multi-model-agent-mcp", "serve"],
+      "args": ["-y", "@zhixuan92/multi-model-agent-mcp", "serve"],
       "env": {
         "OPENAI_API_KEY": "sk-...",
         "ANTHROPIC_API_KEY": "sk-ant-...",
@@ -103,6 +94,20 @@ Add to `claude_desktop_config.json`:
 
 Restart your MCP client after changing config.
 
+## Recommended: delegation rule for Claude Code
+
+Claude Code's native `Task` / `Agent` subagents inherit your parent session's expensive model and eat its context window. We ship a drop-in rule file that teaches Claude Code **when** to delegate work through `delegate_tasks` instead — mechanical edits go to free providers, reasoning-tier work escalates only when needed, and independent tasks run in parallel.
+
+Install globally:
+
+```bash
+mkdir -p ~/.claude/rules
+curl -o ~/.claude/rules/multi-model-delegation.md \
+  https://raw.githubusercontent.com/zhixuan312/multi-model-agent/HEAD/docs/claude-code-delegation-rule.md
+```
+
+Restart Claude Code after installing. The full rule — judgment-vs-labor principle, decision procedure, provider routing table, dispatch shape, verification, escalation ladder — lives at [`docs/claude-code-delegation-rule.md`](https://github.com/zhixuan312/multi-model-agent/blob/HEAD/docs/claude-code-delegation-rule.md). Read that file before adapting it to your own provider names.
+
 ## The `delegate_tasks` tool
 
 Accepts an array of tasks and runs them concurrently. Auto-routes each task by capability filter → quality tier filter → cheapest remaining provider, or pin a task to a specific provider.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@zhixuan92/multi-model-agent-mcp",
-  "version": "0.1.2",
+  "version": "0.1.4",
   "type": "module",
   "license": "MIT",
   "description": "MCP server for multi-model-agent. Exposes a delegate_tasks tool that routes work to Claude, Codex, or OpenAI-compatible sub-agents based on capability, quality tier, and cost.",
@@ -23,7 +23,9 @@
   },
   "homepage": "https://github.com/zhixuan312/multi-model-agent#readme",
   "bugs": "https://github.com/zhixuan312/multi-model-agent/issues",
-  "files": ["dist"],
+  "files": [
+    "dist"
+  ],
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",
   "bin": {
@@ -43,7 +45,9 @@
     "build": "tsc && chmod +x dist/cli.js",
     "prepublishOnly": "npm run build"
   },
-  "engines": { "node": ">=22.0.0" },
+  "engines": {
+    "node": ">=22.0.0"
+  },
   "dependencies": {
     "@zhixuan92/multi-model-agent-core": "^0.1.1",
     "@modelcontextprotocol/sdk": "^1.0.0",