Two providers means two keys and two bills, which is the cost of the combination. What it buys you is the second step: find the paper, then get a critique or a literature-review placement of it without moving the text somewhere else. The five analysis types are genuinely different outputs, not tone settings.
An MCP server that combines two different services: Brave Search for finding things on the web and near a place, and Google's Gemini model for reading an academic paper and reporting on it. The paper analysis is the part you cannot get from a search tool alone.
- General web search with a query up to 400 characters, `count` from 1 to 20 (default 10) and an `offset` up to 9 — `brave_web_search`
- Local search for businesses and places, with `count` from 1 to 20 (default 5) — `brave_local_search`
- A research paper analysed by Gemini — `gemini_research_paper_analysis`, taking the paper text plus an `analysisType` of "summary", "critique", "literature review", "key findings" or "comprehensive" (the default), and optional `additionalContext` to steer it
- Structured analysis output — overview, methodology assessment, key findings, limitations, significance and recommendations
Two keys: a Brave Search API key and a Google API key with access to Gemini models, set as `BRAVE_API_KEY` and `GOOGLE_API_KEY` in a `.env` file or the client's env block. Node.js v18+ recommended. Clone, `npm install`, then `npm run build` (or `npx tsc`) and run `node dist/index.js`. The README recommends raising the client's `timeoutMs` to 120000 to avoid initialization timeouts, and notes that a paper exceeding the model's token limits should be sent as a shorter excerpt.