Apache 2.0

Multi-model orchestration.
Zero manual routing.

Drop-in Claude Code plugin that auto-dispatches tasks across 48+ cataloged models from 6 MCP providers. Opus plans & synthesizes. Claude-native Sonnet/Haiku, MCP providers, and Codex verification run in parallel — silently.

48+
Models
6
MCP catalogs
10
Slash commands
0
Manual routing

How auto-routing works

Every task flows through a four-stage pipeline without you lifting a finger.

1
🧠

Opus Plans

Parses your request, decomposes into subtasks, selects the optimal executor per task signal — silently.

2

Parallel Execution

Sonnet, Haiku, Ollama, NVIDIA, Copilot, Gemini, and opencode run independent subtasks simultaneously in one message.

3
🔍

Codex Verifies

Every non-trivial diff goes through Codex review. Issues are routed back to Sonnet/Haiku for fixes before finishing.

4

Opus Synthesizes

Collects results, resolves conflicts, reports the final outcome — including which route each subtask took.

Everything you need, nothing you don’t

Six design principles that make this plugin actually useful.

🔌

Drop-in Plugin

Two commands and it’s live in any project. The plugin manifest auto-loads all 7 MCP servers. No per-project config files needed.

🤫

Silent Auto-Routing

Opus reads the task signal and picks the right model automatically. It reports the route in one line — then runs. No prompting, ever.

Parallel Execution

Independent subtasks are dispatched in a single message. Multi-file edits, parallel searches, and cross-provider comparisons run concurrently.

🔒

Verification Gate

Codex reviews every non-trivial diff before the task is marked complete. If it flags issues, they’re fixed and re-reviewed automatically.

🌐

Multi-Provider

48+ cataloged models across Ollama cloud, NVIDIA NIM, NVIDIA Security, GitHub Copilot, Google Gemini, and opencode — plus Claude-native executors and Codex verification.

🚀

Zero Configuration

Optional env vars for each provider — nothing is required. Every provider is optional; the plugin degrades gracefully if a key is absent.

Execution lanes: Claude-native + MCP catalogs

The 48+ models span 6 optional MCP catalogs. Claude-native executors and Codex verification are built into the orchestration flow.

Anthropic 3 models
claude-opus-4-7 claude-sonnet-4-6 claude-haiku-4-5
Ollama Cloud 15 models
gemma4:31b-cloud kimi-k2.5:cloud kimi-k2-thinking:cloud glm-5.1:cloud devstral-2:123b-cloud qwen3-coder:480b-cloud +9 more
NVIDIA NIM 10 models
qwen3-coder-480b nemotron-ultra-253b nemotron-super-49b devstral-2-123b kimi-k2-instruct gemma-4-31b llama-405b mistral-large
NVIDIA Security 8 models
nemotron-ultra qwen3-coder devstral llama-guard nemotron-safety nemotron-safety-reason granite-guardian gliner-pii
GitHub Copilot 6 models
gpt-5.3-codex gemini-3-pro gpt-5 claude-sonnet-4.6 claude-opus-4.6 claude-haiku-4.5
Google Gemini 5 models
gemini-3-pro-preview gemini-3-flash-preview gemini-2.5-pro gemini-2.5-flash auto
opencode CLI 4 models
opencode/big-pickle opencode/ling-2.6-flash-free opencode/nemotron-3-super-free opencode/minimax-m2.5-free

Opus routes by task signal

You state the goal. Opus reports the route in one line and runs — no prompting.

Task signal Routed to
Bulk read / grep / rename / format Haiku
Multi-file refactor, debugging, tests Sonnet
Deep chain-of-thought reasoning kimi-k2-thinking:cloud nemotron-ultra
Agentic coding / repo-level edits qwen3-coder glm-5.1:cloud
Coding second opinion / alt-frontier gemma4:31b-cloud nemotron-ultra
Long-context / agentic / vision kimi-k2.5:cloud gemini-3-pro
Multimodal / vision tasks gemma4 (NIM) Gemini auto
Security audit / CVE / OWASP / injection NVIDIA Security
Stuck / failing tests / pre-merge verify Codex
Free-tier bulk / repeat calls opencode CLI
2+ independent subtasks Parallel dispatch

10 commands, full control

All commands are prefixed with /multi-model: inside Claude Code.

/multi-model:delegate <task> Auto-orchestrate any task end-to-end. Opus decomposes, dispatches, verifies, and reports.
/multi-model:ollama <prompt> Direct Ollama cloud chat. Use --model to pick a specific model from the 15-model catalog.
/multi-model:nvidia <prompt> Direct NVIDIA NIM chat. Default: qwen3-coder — best-in-class agentic coding.
/multi-model:nvidia-security <prompt> Security audit, guardrail task, CVE scan, OWASP check, or prompt injection detection.
/multi-model:copilot <prompt> GitHub Copilot CLI. Picks from GPT-5.3-Codex, Gemini 3 Pro, and Anthropic models.
/multi-model:gemini <prompt> Google Gemini CLI — free, no premium cost. Smart routing picks the best Gemini model automatically.
/multi-model:opencode <prompt> opencode CLI with free-tier models only. MCP server enforces the free-models allowlist to prevent billing.
/multi-model:codex [review|rescue|adversarial] <arg> Codex handoff — review diffs, rescue stuck tasks, or run adversarial reviews before merge.
/multi-model:models List all available models across providers with routing descriptions and cost tier.
/multi-model:test-all-models Canary ping every executor and MCP endpoint to verify all providers are healthy.

Up and running in two commands

Works in any project. No global config changes, no per-project boilerplate.

# Add the plugin from the GitHub marketplace (recommended)
claude plugin marketplace add ranjankumarpatel/claude-code-multi-model
claude plugin install multi-model@claude-code-multi-model

# Restart Claude Code — plugin auto-loads 7 MCP servers
# Verify installation:
claude mcp list
# expected: plugin:multi-model:{ollama,nvidia-nim,nvidia-security,copilot,codex,opencode,gemini}

# To update after plugin changes:
claude plugin marketplace update claude-code-multi-model
# Clone for local development / hacking on the plugin itself
git clone https://github.com/ranjankumarpatel/claude-code-multi-model.git
claude plugin marketplace add /absolute/path/to/claude-code-multi-model
claude plugin install multi-model@claude-code-multi-model

# Install plugin dependencies
cd plugins/multi-model
npm install
# Add to your shell profile (~/.bashrc, ~/.zshrc, etc.)
# All optional — omit what you don't need

# NVIDIA NIM + Security (build.nvidia.com)
export NVIDIA_API_KEY="nvapi-..."

# Ollama cloud (default: http://localhost:11434)
export OLLAMA_HOST="http://localhost:11434"

# GitHub Copilot CLI (PAT with Copilot Requests scope)
export GH_TOKEN="ghp_..."

# Google Gemini CLI (or run 'gemini' once for OAuth)
export GEMINI_API_KEY="AIza..."

# Install optional CLIs:
npm install -g @openai/codex         # Codex verification gate
npm install -g opencode-ai           # opencode free-tier
npm install -g @github/copilot       # GitHub Copilot CLI
npm install -g @google/gemini-cli    # Google Gemini CLI

Example — delegate a full feature

# In Claude Code — auto-routed end-to-end
/multi-model:delegate add rate limiting to /api/upload, cover with tests, scan for injection

# Opus reports one-line route then runs:
# Routing: read routes → Haiku | write middleware+tests → Sonnet (parallel)
#          security scan → NVIDIA Security | verify diff → Codex | synthesize → Opus