Gemini Docs MCP Server

What is Gemini Docs MCP Server?

Gemini Docs MCP Server is an MCP server that provides LLMs with access to curated documentation for various technologies using the Gemini API and its 2‑million‑token context window. It is designed for clients like Roo/Cline and replaces web search and traditional RAG by directly feeding the model a complete knowledge base.

How to use Gemini Docs MCP Server?

Install the server via Smithery (npx -y @smithery/cli install @M-Gonzalo/cosa-sai --client claude) or manually configure it in your client’s settings file (e.g., cline_mcp_settings.json). Set the environment variable GEMINI_API_KEY, use bun as the command, and list the tools you want to allow. The server then exposes four tools: can_x_be_done, hints_for_problem, is_this_good_practice, and how_to_do_x.

Key features of Gemini Docs MCP Server

• Uses Gemini API with a 2‑million‑token context window
• Accesses a curated, static knowledge base per technology
• No chunking, vector DB, or custom retriever needed
• Four query tools: can_x_be_done, hints_for_problem, is_this_good_practice, how_to_do_x
• Optional verbose logging for debugging

Use cases of Gemini Docs MCP Server

• Learn an unfamiliar technology by asking natural‑language questions about its documentation
• Get targeted hints for debugging a specific problem in a known stack
• Verify whether a code snippet follows idiomatic practices for the technology
• Discover alternative approaches for a task without searching the web

FAQ from Gemini Docs MCP Server

Does Gemini Docs MCP Server provide real‑time documentation updates?

No. The knowledge base is static and must be manually updated to reflect changes in the technology.

What are the context‑window limitations of Gemini Docs MCP Server?

The server can only see about 2 million tokens at once. For very large technology stacks the entire documentation may not fit in a single request.

What runtime dependencies does Gemini Docs MCP Server require?

It requires a Gemini API key (set as GEMINI_API_KEY), the Bun runtime, and a manually procured documentation knowledge base (e.g., downloaded from a public repo or scraped from a website).

How is the knowledge base obtained and prepared?

You must procure the documentation yourself—by mirroring a website, cloning a repository, etc. An optional sanitation step converts HTML to Markdown and removes styling.

Which LLM model does Gemini Docs MCP Server use?

It uses Gemini 1.5 Pro (not Flash), which makes first queries slower because the documentation must be uploaded to the API.