Installation

Multiple ways to install Ask LLM, depending on whether you want the unified orchestrator or specific providers.

Prerequisites

• Node.js v20.0.0 or higher (LTS 20 or 22)
• An MCP client — Claude Code, Claude Desktop, Cursor, Warp, Copilot, or any of the 40+ compatible clients
• At least one provider CLI installed and authenticated:
  • npm install -g @google/gemini-cli && gemini login for Gemini
  • npm install -g @openai/codex (then follow CLI auth) for Codex
  • Ollama running locally with a model pulled (ollama pull qwen2.5-coder:7b)

Packages

Package	Purpose	Tools exposed
ask-llm-mcp	Unified orchestrator (recommended) — auto-detects all installed providers	ask-llm, multi-llm, get-usage-stats, diagnose, ping
ask-gemini-mcp	Gemini-only — full feature set including @ file syntax, sandbox, edit mode	ask-gemini, ask-gemini-edit, fetch-chunk, get-usage-stats, ping
ask-codex-mcp	Codex-only	ask-codex, get-usage-stats, ping
ask-ollama-mcp	Ollama-only (local)	ask-ollama, get-usage-stats, ping

The unified orchestrator uses a single ask-llm tool with a provider parameter for token efficiency (ADR-029) — you pick the provider per call. Per-provider packages expose the richer per-provider tool surface (ask-gemini-edit for structured edits, fetch-chunk for large response pagination, etc.).

Method 1: NPX (recommended)

No install needed — npx downloads on first invocation and caches:

{
  "mcpServers": {
    "ask-llm": {
      "command": "npx",
      "args": ["-y", "ask-llm-mcp"]
    }
  }
}

For per-provider packages, swap ask-llm-mcp for the package you want.

Method 2: Global install via Claude Code

claude mcp add --scope user ask-llm -- npx -y ask-llm-mcp

# Or per-provider
claude mcp add --scope user gemini -- npx -y ask-gemini-mcp
claude mcp add --scope user codex  -- npx -y ask-codex-mcp
claude mcp add --scope user ollama -- npx -y ask-ollama-mcp

--scope user registers globally for all projects. Drop the flag for per-project scope.

Method 3: Plain npm global

npm install -g ask-llm-mcp
# Binary is at: $(npm bin -g)/ask-llm-mcp

Then point your MCP config at the binary path or just use ask-llm-mcp directly:

{
  "mcpServers": {
    "ask-llm": { "command": "ask-llm-mcp" }
  }
}

Method 4: Claude Code Plugin (richer experience)

If you're a Claude Code user, the plugin adds slash commands (/multi-review, /brainstorm, /compare, /gemini-review, /codex-review, /ollama-review), reviewer subagents with confidence-based filtering, and a pre-commit hook:

/plugin marketplace add Lykhoyda/ask-llm
/plugin install ask-llm@ask-llm-plugins

See Plugin Overview for details.

Verify

# From terminal — works even if MCP isn't configured yet
npx ask-llm-mcp doctor

# From any MCP client — once configured
"Use ask-llm ping to test the connection"

The doctor reports Node version, resolved PATH, every provider CLI's presence + version, and active env vars. Use it as your first stop when something doesn't work.

See Getting Started for the full step-by-step walkthrough or How to Ask for usage patterns.