MCP server with Claude-Gemini collaboration and persistent context management
Server has 8 tools with explicit schemas and descriptions, but quality is inconsistent. Naming is verb-forward but some tools combine multiple concerns (e.g., 'collaborate' does query+exchange+context). Output schemas are not documented. Error handling returns generic success/error objects without recovery guidance. No tool annotations (readOnlyHint, destructiveHint).
Add data to the conversation context for future reference
Clear context data
Start a collaborative conversation between Claude and Gemini with persistent context
Create a new context session
Get a summary of the current context
Get relevant context entries for the current query
Query Gemini directly for a specific response
Missing parameter descriptions on critical fields. 'scope' in add_to_context lacks guidance on valid values beyond enum. 'data' parameter has no type constraint or format guidance.
No output schema documentation. Tools return {success, ...} objects but LLM cannot predict field structure for downstream chaining.
'collaborate' tool combines query + multi-round exchange + context management. Per pattern:tool, each tool should do exactly one thing. Split into separate tools: query_gemini_with_context, continue_collaboration.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | 2026-07-28+ | v2 |
Add tags to existing context data
Error handling returns generic {success: false, error: string} without recovery guidance. Per pattern:recovery-guide, errors must tell LLM what to do next. E.g., 'Session not found. Call create_session() first.'
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). clear_context is destructive but unmarked. Per current MCP spec (2026-07-28), tool annotations enable safer agent planning.