Quick Start

Three steps: install at least one provider CLI, register the MCP server with your client, verify with ping. Start with one provider and add others anytime.

1. Install a provider

Node.js v20+ is required. Then pick a provider. Codex and Claude are the recommended pair: each can review the other.

Codex

npm install -g @openai/codex
# follow the codex CLI's auth instructions

Claude

npm install -g @anthropic-ai/claude-code
# run claude once and authenticate
# (use this provider FROM Codex or another non-Claude host)

Antigravity

# install agy >=1.1.5 from https://antigravity.google, then:
agy --version
# run agy once and authenticate

Ollama

# install from https://ollama.com, then:
ollama pull qwen3.6:27b

Gemini

npm install -g @google/gemini-cli && gemini login
# enterprise seats only since 2026-06-18

Not sure which? See each provider's page: Codex, Claude, Antigravity, Ollama, Gemini.

2. Register the MCP server

The recommended package is @ask-llm/mcp, the unified orchestrator: it auto-detects every provider CLI you installed and exposes one ask-llm tool plus multi-llm, get-usage-stats, diagnose, and ping.

Run in your terminal:

# Project scope (current project only)
claude mcp add ask-llm -- npx -y @ask-llm/mcp

# User scope (all projects)
claude mcp add --scope user ask-llm -- npx -y @ask-llm/mcp

Or install as a plugin (adds slash commands like /multi-review, /brainstorm, /compare, plus reviewer subagents and the opt-in continuous codex-pair review hook):

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

Prefer a single provider with its richer tool surface (ask-codex-edit, fetch-chunk, native session tools)? Install the per-provider package instead: swap @ask-llm/mcp for @ask-llm/codex-mcp, @ask-llm/claude-mcp, @ask-llm/antigravity-mcp, @ask-llm/ollama-mcp, or @ask-llm/gemini-mcp in any tab above.

3. Verify

Ask your agent:

Use ask-llm ping to test the connection

A Pong! reply lists the providers your server detected. If something is off, run the doctor from your terminal; it works even when the MCP server cannot start:

npx @ask-llm/mcp doctor

First calls

Use ask-llm to ask Codex to review the staged changes
Use ask-llm to ask Claude to critique this plan
Use multi-llm to ask Codex and Claude whether this approach is thread-safe

multi-llm returns per-provider responses plus token usage in one structured payload, and one provider hitting quota does not fail the others.

Next steps

• How to Ask: prompt patterns that work
• Multi-Turn Sessions: continue a conversation across calls
• Claude Code Plugin: slash commands, reviewer agents, hooks