Gemini

@ask-llm/gemini-mcpenterprise default gemini-3.1-pro-preview → fallback gemini-3.6-flash

Bridge Claude with Google's Gemini via the official Gemini CLI. Leverages Gemini's massive 1M+ token context window for large file and codebase analysis while Claude handles interaction and code editing.

Discontinued on consumer tiers: migrate to Antigravity

From 2026-06-18, Google restricts Gemini CLI access to Gemini Code Assist Standard/Enterprise seats; free, Google AI Pro, and Ultra accounts lose access. @ask-llm/gemini-mcp still installs and launches; the failure is a runtime auth/quota error from Google's backend, not a missing binary.

On a subscription tier? Migrate to Antigravity (agy), Google's successor CLI, covered by the same AI Pro/Ultra subscription with no per-token billing. Install @ask-llm/antigravity-mcp, or switch to ask-codex / ask-ollama. The 2026-06-18 tier change section below covers what happens at runtime.

2026-06-18 tier change

From 2026-06-18, Google serves Gemini CLI requests only for Gemini Code Assist Standard/Enterprise accounts; free, Google AI Pro, and Ultra accounts lose access. The surviving tiers keep working with the same gemini binary and backend.

What this means for @ask-llm/gemini-mcp:

• The npm package still installs and launches; the binary is unchanged. The failure for non-enterprise accounts is a runtime auth/quota error, not a missing binary, so reinstalling will not help.
• On or after the cutoff, an auth/quota-class failure now surfaces an actionable notice (cutoff date + options) instead of a raw error. The note is advisory ("likely caused by the tier change"); a genuine auth, billing, or quota error can also trigger it.
• Google's successor is the Antigravity CLI (agy), a separate, closed-source binary that @ask-llm/gemini-mcp does not wrap. Use the dedicated @ask-llm/antigravity-mcp package (covered by the same Google AI Pro/Ultra subscription, no per-token billing), run agy directly, or switch to ask-codex / ask-ollama.
• Testing the guidance: set ASK_GEMINI_TIER_CUTOFF to a past UTC instant (e.g. 2020-01-01T00:00:00Z) to force the post-cutoff gate on; an auth/quota failure will then prepend the notice. The default cutoff is 2026-06-18T00:00:00Z.

Google's announcement.

Best for: whole-codebase reads using the 1M+ token context window, if you have an eligible Gemini Code Assist Standard/Enterprise seat. Not for: most users from 2026-06-18 (see the notice above). For large-context reads without an enterprise seat, use Antigravity.

Installation

1. Provider CLI

npm install -g @google/gemini-cli && gemini login

2. Register the MCP server (Claude Code shown; see Quick Start for other clients)

claude mcp add --scope user gemini -- npx -y @ask-llm/gemini-mcp

Or install globally: npm install -g @ask-llm/gemini-mcp

Prerequisites

1. Node.js v20.0.0 or higher
2. Gemini CLI installed and authenticated:

npm install -g @google/gemini-cli
gemini login

Tools

Tool	Purpose
ask-gemini	Send prompts to Gemini CLI with @ file syntax. Optional sessionId for multi-turn; live progressive output via --output-format stream-json
ask-gemini-edit	Structured code edits via Gemini changeMode. Returns OLD/NEW edit blocks. Supports includeDirs for monorepo context
fetch-chunk	Retrieve subsequent chunks from cached large responses
get-usage-stats	Per-session token totals + breakdowns by provider/model. In-memory
ping	Fast connection test to verify MCP setup

ask-gemini returns both human-readable text and a structured AskResponse (provider, response, model, sessionId, usage) via MCP outputSchema; programmatic clients can extract the sessionId and usage fields directly without parsing the response footer.

Models

When the default model hits quota, the executor retries once on the fallback and reports the actual model used in the response.

• Default: gemini-3.1-pro-preview (latest, highest capability)
• Quota fallback: gemini-3.6-flash, automatic on RESOURCE_EXHAUSTED
• Overrides: ASK_GEMINI_MODEL, ASK_GEMINI_FALLBACK_MODEL, or the per-call model parameter

Key Features

• 1M+ token context for analyzing entire codebases
• Multi-turn sessions via sessionId: native --resume <id> (zero replay cost)
• Include directories for monorepo context (includeDirs parameter on ask-gemini-edit)
• Live progressive output: assistant message deltas stream to MCP progress notifications, no frozen waits on long calls
• Structured AskResponse via outputSchema for programmatic clients
• Automatic quota fallback from Pro to Flash on RESOURCE_EXHAUSTED

npm

• Package: @ask-llm/gemini-mcp
• Binary: ask-gemini-mcp