Google Drive MCP Server

Exposes Google Drive's native search API through the Model Context Protocol's tool interface, allowing LLM clients to query files by name, metadata, and MIME type without direct API credentials. Implements MCP's tool registration pattern to translate natural language search intents into Google Drive query syntax, handling pagination and result filtering server-side before returning structured file metadata to the client.

Retrieves the full text content of Google Docs documents through the Google Drive API, converting Google's proprietary document format into plain text or structured markdown while preserving document hierarchy (headings, lists, tables). Implements streaming content retrieval to handle large documents efficiently, with server-side caching to reduce redundant API calls for frequently accessed documents.

Retrieves data from Google Sheets as structured JSON or CSV, automatically inferring column types and handling sparse data. Uses the Google Sheets API to fetch cell ranges with formatting metadata, then transforms the raw grid into columnar format with optional type coercion (dates, numbers, booleans). Supports filtering and sorting server-side to reduce payload size for large sheets.

Extracts text content, speaker notes, and slide structure from Google Slides presentations via the Google Slides API, converting slide hierarchy into a traversable outline format. Handles both text-based content and metadata (slide titles, notes, layout information) while optionally exporting slide images as base64-encoded data for multimodal LLM processing. Implements lazy loading to avoid fetching all slides upfront.

Lists contents of a Google Drive folder with optional recursive traversal to build a complete folder tree structure. Uses the Google Drive API's file list endpoint with parent folder filtering, implementing depth-first or breadth-first traversal patterns to map folder hierarchies. Returns structured metadata for each file/folder (id, name, mimeType, size, permissions) enabling clients to understand Drive organization without manual navigation.

Exposes Google Drive files as MCP resources, allowing LLM clients to reference files by URI (e.g., 'gdrive://file-id') and retrieve their metadata or content on-demand. Implements MCP's resource protocol to provide file handles that clients can pass between tools, enabling workflows where one tool's output (a file ID) becomes another tool's input without explicit serialization. Supports resource templates for common patterns (e.g., 'gdrive://folder/{folder-id}/files').

Manages Google OAuth 2.0 authentication for the MCP server, handling credential storage, token refresh, and expiration. Implements automatic token refresh before expiration to prevent mid-request failures, with fallback to credential re-authentication if refresh fails. Supports both service account credentials (for server-to-server access) and user credentials (for user-delegated access), with secure credential storage using environment variables or encrypted local storage.

Implements resilient error handling for Google Drive API calls with exponential backoff retry logic for transient failures (rate limits, timeouts, temporary service errors). Distinguishes between retryable errors (429, 500, 503) and permanent failures (401, 403, 404) to avoid wasting retries on unrecoverable errors. Tracks API quota usage and implements client-side rate limiting to prevent hitting Google's quotas, with configurable backoff strategies.

Automatically generates MCP tool schemas that describe available Google Drive operations (search, read, list) with input parameters, output types, and descriptions. Uses introspection of the Google Drive API to build tool definitions that MCP clients can discover and invoke, with parameter validation and type coercion. Implements schema versioning to support API changes without breaking clients.

Enables batch operations on multiple Google Drive files (e.g., bulk read, bulk metadata update) with transactional guarantees where possible. Implements request batching to reduce API call overhead, grouping multiple operations into single batch requests. Provides rollback semantics for write operations: if any operation in a batch fails, all changes are reverted (where supported by Drive API).