Adding native support for Google Gemini Embeddings

[josepcrespo]

Current Problem / 当前问题
Google Gemini is a high-quality embedding provider with a generous free tier (100 RPM via the OpenAI-compatible endpoint), but it is not natively supported in MCPHub. Users who want to use Google Gemini must manually discover and enter the correct base URL, model name, and configuration, with no guidance from the UI or the MCPHub documentation.

Although Google Gemini works via the OpenAI-compatible endpoint, it is not listed as a known provider, not surfaced in any dropdown or preset, and its availability and free-tier status are not communicated to users anywhere in the application.

Proposed Solution / 建议方案
Add Google Gemini as a first-class supported embedding provider in MCPHub:

• Include it in the predefined provider list (see related feature request Predefined Embeddings Provider List with Autocomplete #636) with pre-filled defaults:
  • Base URL: https://generativelanguage.googleapis.com/v1beta/openai
  • Default Model: gemini-embedding-001
  • Free Tier: Yes — 100 RPM
• Ensure the runtime encoding format detection (see related feature request Embeddings Encoding Format Detection: Replace Static Provider List with Runtime Detection #634) correctly handles Google Gemini's response format.
• Document supported embedding dimensions for gemini-embedding-001: 128, 256, 512, 768, 1536, 2048, and 3072 (auto-normalized).
• Note in the documentation that for dimensions other than 3072, manual normalization is required for accurate cosine similarity results.

Alternatives / 替代方案

• Document Google Gemini as a supported "OpenAI-compatible" provider in the documentation without changes to the UI, relying on users to configure it manually.
• Add Google Gemini support using the native Google AI SDK instead of the OpenAI-compatible endpoint, which would enable access to Gemini-specific features not available through the compatibility layer.

Additional Context / 补充信息
Google Gemini (gemini-embedding-001) has been tested and confirmed to work correctly with the current MCPHub embedding system via the OpenAI-compatible endpoint, including with auto encoding format mode. The free tier (100 RPM) is particularly attractive for self-hosted or low-traffic deployments where users want high-quality embeddings without incurring cost. Highlighting this option prominently in the UI would lower the barrier to entry for new users evaluating MCPHub.