npm · @qmediat.io/gemini-code-context-mcp
$npx -y @qmediat.io/gemini-code-context-mcp@1.20.0 GEMINI_CREDENTIALS_PROFILE — Profile name in ~/.config/qmediat/credentials (chmod 0600). Created by `npx @qmediat.io/gemini-code-context-mcp init`. Recommended; keeps your API key out of ~/.claude.json.
GEMINI_API_KEY · secret — Fallback Tier 3 auth. Your Gemini API key. Emits a warning at startup recommending you move it to the credentials profile via the init command.
GEMINI_USE_VERTEX — Set to `true` to use Vertex AI backend via Application Default Credentials. Requires GOOGLE_CLOUD_PROJECT.
GOOGLE_CLOUD_PROJECT — GCP project ID when using Vertex AI backend. Only read when GEMINI_USE_VERTEX=true.
GEMINI_DAILY_BUDGET_USD — Hard daily USD cap enforced locally. Server refuses calls after the cap until UTC midnight. Unlimited if unset. Honoured by `ask`, `code`, and per-iteration by `ask_agentic`.
GEMINI_CODE_CONTEXT_DEFAULT_MODEL — Model alias (`latest-pro-thinking`, `latest-pro`, `latest-flash`, `latest-lite`, `latest-vision`) or literal model ID, read by ask, ask_agentic and code. Default: `latest-pro-thinking` (the thinking tier). code replaces a configured default that resolves but cannot reason or think with `latest-pro-thinking` and reports it as `configuredModelReplaced`; a default that does not resolve fails every tool.
GEMINI_CODE_CONTEXT_CACHING_MODE — `implicit` (default since v1.14.0: the workspace text is sent inline, Gemini's automatic prefix cache decides any discount) or `explicit` (Files API upload + Context Cache with Google's cached-input price). Per-call `cachingMode` wins.
GEMINI_CODE_CONTEXT_CACHE_TTL_SECONDS — Context Cache TTL in seconds in explicit caching mode. Default: 3600 (1 hour). Hot workspaces (<10 min since last use) auto-refresh via background watcher.
GEMINI_CODE_CONTEXT_LOG_LEVEL — `debug` | `info` | `warn` | `error`. Default: `info`.