@zhixuan92/multi-model-agent-mcp 2.7.5 → 2.8.1

package/README.md

@@ -1,143 +1,21 @@
-# @zhixuan92/multi-model-agent-mcp
+# @zhixuan92/multi-model-agent-mcp — DEPRECATED
 
-**MCP server for multi-model-agent.** Your AI assistant gets 10 tools for delegating work to cheaper agents — parallel execution, cross-agent review, 90% cost savings.
+This package has been replaced by **[`@zhixuan92/multi-model-agent`](https://www.npmjs.com/package/@zhixuan92/multi-model-agent)** in 3.0.0.
 
-Works with Claude Code, Codex CLI, Cursor, Gemini CLI, and Claude Desktop.
+The 3.0.0 release removes all MCP server support. Use the new standalone HTTP service + client-installable skills instead.
 
-## Install
-
-Requires Node >= 22 and a config file at `~/.multi-model/config.json`.
-
-**1. Create config** — define your two agent slots. Three agent types are supported:
-
-| Type | Auth | API key needed? |
-|---|---|---|
-| `claude` | Your existing Claude Code / Claude subscription | No — uses local OAuth |
-| `codex` | Your existing Codex subscription (`codex login`) | No — reads `~/.codex/auth.json` |
-| `openai-compatible` | Any OpenAI-compatible API (GPT, MiniMax, DeepSeek, Groq, local vLLM) | Yes — `apiKeyEnv` or `apiKey` |
-
-**Example — Claude + Codex (no API keys):**
+## Migration
 
 ```bash
-mkdir -p ~/.multi-model && cat > ~/.multi-model/config.json << 'EOF'
-{
-  "agents": {
-    "standard": { "type": "codex", "model": "codex-mini-latest" },
-    "complex": { "type": "claude", "model": "claude-sonnet-4-20250514" }
-  },
-  "defaults": { "timeoutMs": 1800000, "maxCostUSD": 10, "tools": "full" }
-}
-EOF
-```
+npm uninstall -g @zhixuan92/multi-model-agent-mcp
+npm install -g @zhixuan92/multi-model-agent
 
-**Example — OpenAI-compatible endpoints (API keys required):**
-
-```bash
-mkdir -p ~/.multi-model && cat > ~/.multi-model/config.json << 'EOF'
-{
-  "agents": {
-    "standard": {
-      "type": "openai-compatible",
-      "model": "MiniMax-M2",
-      "baseUrl": "https://api.minimax.io/v1",
-      "apiKeyEnv": "MINIMAX_API_KEY"
-    },
-    "complex": {
-      "type": "openai-compatible",
-      "model": "gpt-5",
-      "baseUrl": "https://api.openai.com/v1",
-      "apiKeyEnv": "OPENAI_API_KEY"
-    }
-  },
-  "defaults": { "timeoutMs": 1800000, "maxCostUSD": 10, "tools": "full", "parentModel": "claude-opus-4-6" }
-}
-EOF
+mmagent serve              # start the daemon
+mmagent install-skill      # install skills for your AI client
 ```
 
-> **`parentModel`** (optional): When set, headlines show `$Y saved vs model (Zx ROI)`. When omitted, headlines show `$X actual`.
-
-Mix and match freely — e.g., `claude` for complex + `openai-compatible` for standard.
-
-**2. Register the MCP server:**
-
-```bash
-# Claude/Codex agents (no env vars needed):
-claude mcp add multi-model-agent -s user \
-  -- npx -y @zhixuan92/multi-model-agent-mcp serve
-
-# OpenAI-compatible agents (pass API keys):
-claude mcp add multi-model-agent -s user \
-  -e MINIMAX_API_KEY=... -e OPENAI_API_KEY=... \
-  -- npx -y @zhixuan92/multi-model-agent-mcp serve
-```
-
-No install step, no long-running process. Your MCP client spawns it on demand via `npx`.
-
-For Codex CLI, Claude Desktop, and Cursor setup, see the [full guide](https://github.com/zhixuan312/multi-model-agent#quick-start).
-
-## What you get
-
-| Tool | What it does |
-|---|---|
-| `delegate_tasks` | Dispatch tasks in parallel with minimal input: `prompt` plus optional `agentType`, `filePaths`, `done`, and `contextBlockIds`. The MCP interprets your request and infers missing details — if confused, it returns a proposed interpretation for confirmation. |
-| `audit_document` | Audit docs/files for issues — parallel per file. Accepts `contextBlockIds` for delta audits (round 2+). |
-| `review_code` | Code review with spec + quality pipeline — parallel per file. Accepts `contextBlockIds` for diff-scoped/delta reviews. |
-| `verify_work` | Verify work against a checklist — parallel per file. Accepts `contextBlockIds` for shared context. |
-| `debug_task` | Hypothesis-driven debugging with file context. Accepts `contextBlockIds` for shared context. |
-| `execute_plan` | Execute tasks from a plan document — worker reads plan files, finds matching task by descriptor, implements it. Multiple tasks run in parallel. |
-| `register_context_block` | Store reusable context for later tasks |
-| `retry_tasks` | Re-run specific tasks from a previous batch |
-| `get_batch_slice` | Fetch output or telemetry from a previous batch |
-| `confirm_clarifications` | Resume a clarification set by confirming or editing proposed interpretations |
-
-## Diagnostic logging
-
-Diagnostic logging is OFF by default.
-
-It stays disabled when the `diagnostics` block is absent or when `diagnostics.log` is `false` in `~/.multi-model/config.json`.
-
-To capture a crash/disconnect log to send us, add a `diagnostics` block to your config.
-
-Minimal example:
-
-```json
-{
-  "diagnostics": { "log": true }
-}
-```
-
-Full config shape example:
-
-```json
-{
-  "agents": {
-    "standard": { "type": "codex", "model": "codex-mini-latest" },
-    "complex": { "type": "claude", "model": "claude-sonnet-4-20250514" }
-  },
-  "defaults": {
-    "timeoutMs": 1800000,
-    "maxCostUSD": 10,
-    "tools": "full"
-  },
-  "diagnostics": {
-    "log": true,
-    "logDir": "/some/path"
-  }
-}
-```
-
-`diagnostics.logDir` is optional; when omitted, logs default to `~/.multi-model/logs/`.
-
-When enabled, the server appends JSONL records to `mcp-YYYY-MM-DD.jsonl` in append mode.
-
-Only crash/disconnect diagnostic events are logged: `startup`, `request_start`, `request_complete`, `shutdown`, and `error`. This is a crash-diagnosis log, not a progress feed.
-
-## Setup & Configuration
-
-See the full setup guide with config examples, client-specific instructions, and auth details:
-
-→ **[github.com/zhixuan312/multi-model-agent](https://github.com/zhixuan312/multi-model-agent#quick-start)**
+See the [3.0.0 CHANGELOG](https://github.com/zhixuan312/multi-model-agent/blob/master/CHANGELOG.md#300) for details.
 
-## License
+## About this version (2.8.1)
 
-[MIT](./LICENSE) — Copyright (c) 2026 Zhang Zhixuan
+`2.8.1` ships a 10-line stub that prints a migration notice and exits with code 1. It does not function as an MCP server. Installing or invoking `mmagent` / `multi-model-agent` from this package will redirect users to the new package.