Multi-AI MCP bridge: Gemini + OpenRouter (400+ models) — text, code, image, video, TTS, RAG, Deep Research
This server shows moderate definition quality with significant gaps. 21 tools are registered with descriptions and parameter schemas, but the quality is inconsistent. Most tool descriptions are adequate (100-300 chars), but many parameters lack proper type constraints, defaults, or validation guidance. Error handling is minimal, no recovery guidance, no retryability classification, no actionable error messages visible in the code. The server exposes many complex AI model interfaces (Gemini, OpenRouter, RAG, Deep Research) but abstracts away internal details without clear examples of how agents should compose them. Input validation exists (secure_read_file, validate_path) but is not reflected in parameter constraints or error messages. Critically, tools like `gemini_analyze_codebase`, `gemini_deep_research`, `gemini_generate_image`, and `gemini_generate_video` accept paths and model selections but provide no guidance on format, validation rules, or recovery. Several tools (13-17, 19-21) have minimal or inferred descriptions in the provided code excerpt. The server lacks tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite having clearly destructive (delete_conversation) and read-only operations.
Ask Gemini a question or submit a prompt for text generation.
Query any model via OpenRouter or Gemini API.
Create a file store for RAG operations.
Delete a conversation from storage.
Analyze large codebases using Gemini's 1M token context window. Perfect for architecture analysis, cross-file review, and understanding complex projects.
Analyze images using Gemini vision capabilities. Describe, extract text (OCR), identify objects, or answer questions about images.
9 of 21 tools (43%) have minimal or missing parameter descriptions in the provided code (tools 13-21). The excerpt shows only imports and no detailed docstrings for these tools. Cannot verify schema quality for these tools.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are visible. Tools like `delete_conversation` (destructive) should be annotated to guide agent behavior. This is a current protocol feature (2026-07-28) that should be adopted.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 50 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 57 | - | v1 |
Brainstorm ideas using Gemini with various methodologies.
Challenge and critique an idea, plan, or piece of code from multiple perspectives.
Review code for quality, security, performance, and maintainability issues.
Execute comprehensive research using Google's Deep Research Agent. Autonomously conducts multi-step web searches and synthesizes findings.
Search through files using Gemini's RAG capabilities with semantic understanding.
Generate code in multiple programming languages with optional context from existing files.
Generate images using Gemini/Imagen native image generation. Use descriptive prompts (not keywords) for best results.
Generate videos using Google Veo 3.1 with native audio. Creates 4-8 second videos with realistic motion, dialogue, and sound effects.
List all available Gemini models.
Convert text to speech with 30 voice options. Supports single and multi-speaker (up to 2) audio.
Transcribe audio files using Gemini's speech-to-text capabilities.
Search the web using Gemini with Google Search grounding. Returns answers with citations from authoritative sources.
List all stored conversations.
List all file stores available for RAG operations.
Upload a file to a file store for RAG operations.
Error handling is not evident in visible code. No recovery guidance, no retryability classification, no actionable error messages. Tools like `gemini_analyze_codebase` and `gemini_deep_research` have long-running operations but no timeout or retry logic visible.
Parameter 'continuation_id' in multiple tools (gemini_analyze_codebase, gemini_deep_research, ask_gemini) lacks validation guidance. Should document: format, max length, when it's required vs. optional, and what happens if invalid.
Many tools accept file paths (analyze_codebase, generate_code, generate_image output_path). Descriptions mention 'paths' and 'glob patterns' but do not specify: allowed directories, path traversal protection, symlink handling, or what errors occur with invalid paths.
Tools returning generated content (generate_image, generate_video, text_to_speech) do not document the return format when output_path is not provided. Are they base64-encoded? Raw binary? This ambiguity forces agents to guess.
Models are exposed as enums (flash, pro, fast, flash-lite, veo31, veo31_fast, veo31_lite) but parameters do not explain trade-offs (speed, cost, quality). Agents cannot reason about which to choose without external knowledge.
No visible rate limiting or quota guidance. Tools like `gemini_deep_research` (max_wait_minutes: 5-60) and batch file search could exhaust quotas. Descriptions should warn about rate limits and suggest batching strategies.