Backend for MCPStudio - The Postman for Model Context Protocol. Provides REST and WebSocket APIs for managing MCP servers, tools, and their execution.
MCPstudio Backend exposes 4 Google Drive tools via FastAPI HTTP. Tool definitions are present with basic schemas and descriptions, but fall short of production quality. Naming follows verb_noun convention (listFiles, getFileContent, searchFiles, createFolder), but descriptions are minimal (20-30 chars), lack LLM-optimized context for selection and prerequisites, and parameter descriptions are sparse. Input schemas exist but lack comprehensive validation guidance. No error handling responses are visible in the tool definitions, the control layer (tool_controller.py) shows generic 500 errors with no recovery guidance. Output schemas are not documented, making it impossible for LLMs to predict response structures and plan multi-step sequences. Risk annotations exist (READ_ONLY, WRITE) but toolAnnotations are not formalized in the MCP layer. The server operates as a managed backend with FastAPI endpoints rather than a true MCP-compliant tool server, tool definitions appear to be synthesized from Google Drive API wrappers rather than explicitly registered MCP tools with full schemas.
Creates a new folder in Google Drive
Retrieves the content of a file from Google Drive
Lists files in a Google Drive folder
Searches for files in Google Drive
Tool descriptions are under 30 characters and lack LLM-optimized context. 'Lists files in a Google Drive folder' (34 chars) does not explain WHEN to call listFiles vs searchFiles, what parameters are recommended, or what fields the response contains. LLMs cannot distinguish tools or plan follow-up calls.
Output schemas are completely undocumented. No response type definitions visible for any tool. LLMs cannot predict what fields are returned (e.g., does listFiles return file_id and name, or raw API objects?), blocking multi-step reasoning and forcing agents to use generic field extraction.
No error handling guidance in tool definitions. tool_controller.py catches all exceptions and returns generic HTTP 500 with str(e). LLMs receive no actionable recovery hints: is the folder ID invalid? Does it need a lookup first? Is the service temporarily down?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | 2026-07-28+ | v2 |
Parameter descriptions are minimal or missing rationale. 'ID of the folder to list files from (optional, defaults to root)' for folderId lacks format guidance (is it a Google Drive share ID? An opaque string?). 'Maximum number of files to return' (maxResults) does not explain the default (30) or limits (1-1000?).
listFiles and searchFiles both operate on Google Drive files but serve overlapping purposes. Tool selection guidance is absent, when should an LLM call listFiles vs searchFiles? Does listFiles only work on a single folder? Does searchFiles include folder contents recursively?
No pagination support visible. searchFiles and listFiles accept a maxResults limit but no cursor or next_token field documented. If a folder has 500+ files, agents cannot fetch beyond the first 30. Large result sets will either truncate silently or blow context windows.
createFolder accepts optional parentId but lacks guidance on what happens if parentId is invalid or inaccessible. Does the call fail? Does it default to root? Are there permission errors? Error handling is silent.