shariqriazz/vertex-ai-mcp-server

bun install
bun run build
bunx vertex-ai-mcp-server

Usage

• •Set AI_PROVIDER to 'vertex' (requires GOOGLE_CLOUD_PROJECT and auth, e.g. gcloud auth application-default login) or 'gemini' (requires GEMINI_API_KEY).
• •Run standalone via bunx vertex-ai-mcp-server, install globally with bun install -g vertex-ai-mcp-server, or install for Claude Desktop via Smithery (bunx -y @smithery/cli install @shariqriazz/vertex-ai-mcp-server --client claude).
• •Configure with Cline by pointing the command at build/index.js or using bunx in .roo/mcp.json.
• •Tune VERTEX_MODEL_ID/GEMINI_MODEL_ID, temperature, streaming, max output tokens, retry delay, and retries via env vars.

Key features

• ✓Web-search-grounded answers (answer_query_websearch) and direct knowledge answers (answer_query_direct).
• ✓Research tools: code analysis with docs, technical comparison, architecture pattern recommendation, dependency vulnerability scan, database schema analyzer, security/compliance/testing advisors.
• ✓Filesystem and terminal-execution tools: read/write/edit/list/move/search files and execute_terminal_command.
• ✓Combined AI + filesystem tools (save_generate_project_guidelines, save_doc_snippet, etc.).
• ✓Streaming API by default, basic retry logic, and configurable model/temperature.

Best for

Developers who want Gemini-based coding assistance, doc research, and repo/file operations inside MCP clients like Cline.

Caveats

• ⚠README explicitly warns it includes filesystem-write and terminal-execution tools: run with a dedicated low-privilege identity, limit access, and keep consequential tools behind explicit approval.
• ⚠Requires a GCP project with billing and the Vertex AI API enabled, or a GEMINI_API_KEY for provider 'gemini'.
• ⚠Minimal safety filters (BLOCK_NONE) are applied to reduce blocking — use with caution.
• ⚠MIT; requires Node.js v18+ and Bun.