MCP server for Google Drive file operations and management
This server has moderate structural issues. Tool naming is reasonable (verb-prefix convention followed), but descriptions are superficial and parameter documentation is inconsistent. Most critically, parameter descriptions lack detail about constraints, format expectations, and valid values. Input schemas are present but minimal, they specify types but omit crucial metadata like min/max bounds, regex patterns, and enums. Output schemas are completely undocumented, responses are human-readable formatted strings rather than structured objects, making it hard for LLMs to extract and chain data. Error handling is basic (formatted strings) with no recovery guidance. The server lacks output schema documentation (critical gap), parameter validation metadata, and composition support (no chaining IDs, no pagination). Security is adequate (credentials via env vars) but missing audit logging and permission declarations.
Create a new folder in Google Drive
Download file content from Google Drive
Get detailed metadata about a file
List files in Google Drive
Search for files in Google Drive
Upload a text file to Google Drive
Output schemas completely undocumented. All tools return free-form formatted strings instead of structured JSON. LLMs cannot extract fields for downstream chaining (e.g., file IDs, folder paths). This breaks the tool-chain pattern and forces expensive re-parsing by the agent.
Parameter descriptions lack constraint metadata. 'pageSize' description says 'Default is 10, maximum is 100' but no minimum is stated, no validation logic visible in code (code clamps to [1,100] but description doesn't mention minimum=1). Descriptions for folderName, fileName, content lack format/length guidance. No regex patterns, length limits, or character restrictions documented.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 37 | <=2025-11-25 | v2 |
Responses do not return structured data needed for chaining. listFiles and searchFiles return formatted strings with file IDs embedded in text, not in a structured 'files' array with 'id' fields. createFolder returns success message, not the created folder's ID. downloadFile returns content as string, not {content, mimeType, size}. Downstream tools cannot easily extract IDs to construct subsequent calls.
No pagination support. listFiles and searchFiles accept pageSize but no cursor, offset, or next_page_token parameter. Large result sets will truncate without the LLM knowing more items exist. Code applies orderBy and pageSize correctly in API calls, but tool interface doesn't expose continuation.
Error responses provide no recovery guidance. formatError() returns '❌ Error: <message>' with no next steps. When getDriveClient() fails due to missing credentials, the message says what's missing but doesn't guide the user/agent on how to set environment variables or validate configuration.
Tool descriptions are generic and brief (10-60 characters). E.g., 'List files in Google Drive' (23 chars) lacks context on when to use it vs searchFiles, what data structure is returned, or when folder listing fails. Production baselines show 50-200 chars for optimal LLM disambiguation.
No idempotency guarantees documented. createFolder and uploadFile are write operations that could have side effects on retry, but tool descriptions do not declare whether they are idempotent or require a confirmation step. Code does not implement request deduplication via idempotency keys.
No tool annotations present. Tools lack readOnlyHint, destructiveHint, or idempotentHint metadata. MCP spec 2026-07-28 recommends annotating write/delete tools so agents can apply caution. No annotations visible in code.
No permission or scope declarations. Tools do not declare what Google Drive permissions they require (e.g., 'drive.file' for read, 'drive' for write). This prevents least-privilege agent configuration and clear audit trails.