LabscoMCP early access ↗
RLabs-IncMCP SERVER

Gemini MCP Server by RLabs

Put Gemini 3 — with grounded search, deep research, media generation and code execution — behind one connection.

by RLabs-Inc ↗Model Routing, Multi-Model Consultation & Cost ControlCommunity

LABSCO SUMMARY

Reach for it when your own model cannot do the thing at all.

The tools worth connecting this for are the ones that are not substitutable: grounded search with citations, a research agent that runs for several steps, image and video generation, and Python that actually runs rather than being written and hoped over. The context cache is the detail that keeps it affordable — put a large document in once, query it repeatedly, and stop paying to resend it. Count tokens first if the document is big.

What it is

A server that gives one assistant access to another. It exposes Gemini's model capabilities as tools, so Claude or any MCP client can hand work to Gemini — search that is grounded in live results, research that runs for several steps, media generation, and Python that actually executes.

What you get

  • Ask Gemini directly with control over how hard it thinks — gemini-query, with gemini-brainstorm for a back-and-forth between the two models
  • Grounded answers — gemini-search returns live web information with inline citations, and gemini-deep-research runs an autonomous multi-step investigation, followed with gemini-check-research and gemini-research-followup
  • Read what is on the web — gemini-analyze-url, gemini-compare-urls and gemini-extract-from-url
  • Video and documents — gemini-youtube and gemini-youtube-summary with timestamp clipping, gemini-analyze-document, gemini-summarize-pdf and gemini-extract-tables
  • Images — gemini-generate-image up to 4K across ten aspect ratios, then an editing session you refine over several turns with gemini-start-image-edit, gemini-continue-image-edit, gemini-list-image-sessions and gemini-end-image-edit; gemini-image-prompt helps write the prompt and gemini-analyze-image reads one back
  • Video generation with Veo — gemini-generate-video starts it, gemini-check-video polls and downloads when it is done
  • Speech — gemini-speak and gemini-dialogue for two speakers, across thirty voices, with gemini-list-voices
  • gemini-run-code, where Gemini writes Python and runs it with pandas, numpy and matplotlib available
  • Structured results — gemini-structured validates against a schema you supply, gemini-extract pulls entities, facts and sentiment out of text
  • Cost control — gemini-count-tokens before you spend, and a cache you manage yourself with gemini-create-cache, gemini-query-cache, gemini-list-caches and gemini-delete-cache so a large document is not re-sent every turn
  • A gcli command-line tool ships in the same package, so the same capabilities work outside an MCP client
  • gemini-analyze-code and gemini-analyze-text for a second opinion on quality, security or performance

Requirements

A Google Gemini API key in GEMINI_API_KEY. Install with npx or bunx @rlabs-inc/gemini-mcp, or globally for the gcli command, which stores the key for you. GEMINI_OUTPUT_DIR sets where generated images, videos and audio are written, and VERBOSE turns on logging. You pay Google for what Gemini does; the server adds nothing on top.

Setup effort

One command plus a key — npm install -g @rlabs-inc/gemini-mcp, then supply credentials

See installation options