vertex-rag-mcp

MCP server exposing Vertex AI RAG mini_rag.

Tools

• mini_rag(query, drive_id, top_k=50) — query Vertex RAG against a Drive folder.
• list_drive_files(drive_id, recursive=True, include_folders=False, limit=0) — list Drive files in a folder with sizes and timestamps.
• read_multiple_text_files(drive_id, file_ids=None, include_types=None, all_files=False, recursive=True, output_format="markdown", max_files=0, max_total_chars=0, max_file_chars=0, max_pages=0, max_table_rows=50, max_table_cols=12, chunk_size=0) — extract text from documents and return Markdown/text. CSV is rendered as a Markdown table (limited). Each file starts with a YAML-style metadata block (id, name, url, owner, size, created, lastModified).
• read_multiple_md_files(drive_id, file_ids=None, include_types=None, all_files=False, recursive=True, convert_pdf_to_gdoc=True, cleanup_converted=True, strip_images=True, max_files=0, max_total_chars=0, max_file_chars=0, max_pages=0, max_table_rows=50, max_table_cols=12, chunk_size=0) — extract text as Markdown, preferring Google Docs export. PDFs can be converted to Docs first. strip_images removes embedded data:image blocks.
• export(drive_id) — export Drive markdown to ./output/{drive_id}-vertex-rag.md.
• read_multiple_binary_files(drive_id, file_ids=None, include_types=None, all_files=False, recursive=True, tool_result="resource_link", response_format="url", max_files=0, max_total_bytes=0, max_file_bytes=0, chunk_size=0) — list Drive documents with content for indexing (defaults exclude media). tool_result controls content[] blocks (resource_link vs resource), response_format controls structuredContent (url vs b64_json). Limits are optional; use 0 to disable.
• read_single_file_raw(file_id, export_mime=None, chunk_size=0) — return raw base64 content for a single file (content only, no structuredContent).
• incremental_update_corpus(drive_id, delete_removed=False, dry_run=False) — update a Drive-backed corpus by file modified times.
• full_refresh_corpus(drive_id, delete_old=True) — rebuild a corpus from scratch and optionally delete previous versions.

Default indexable MIME types (no media)

• application/pdf
• application/rtf
• application/vnd.google-apps.document
• application/vnd.google-apps.presentation
• application/vnd.google-apps.spreadsheet
• application/vnd.oasis.opendocument.text
• application/vnd.openxmlformats-officedocument.presentationml.presentation
• application/vnd.openxmlformats-officedocument.spreadsheetml.sheet
• application/vnd.openxmlformats-officedocument.wordprocessingml.document
• text/csv
• text/html
• text/markdown
• text/plain

Setup

• Create fastagent.secrets.yaml in the repo root (see fastagent.secrets.yaml.example).
• Ensure Google credentials are available (for example GOOGLE_APPLICATION_CREDENTIALS).
• Grant the service account access to the Google Drive folder you query.
• read_multiple_md_files with convert_pdf_to_gdoc=True requires Drive write access to create/delete temporary Google Docs.
• Optional: set VERTEX_RAG_MCP_CONVERT_SHARED_DRIVE_ID to a Shared Drive ID to store temporary conversion Docs there (recommended for service accounts).
• Optional: set VERTEX_RAG_MCP_SKIP_OWNERS to a comma-separated list of owner identifiers to skip during indexing/export (matches owner email or display name).

Run

python -m vertex_rag_mcp.server

If installed as a package:

vertex-rag-mcp