Gemini Researcher
Proxy heavy research questions to Gemini CLI through MCP so your main coding agent spends fewer tokens and stays stateless.
Overview
io.github.capyBearista/gemini-researcher is a MCP server for the Idea phase that proxies research queries to Gemini CLI to cut primary-agent context use.
What is this MCP server?
- Stateless MCP server proxying research queries to Gemini CLI
- npm package gemini-researcher with stdio transport (version 1.0.2)
- Optional GEMINI_API_KEY or existing Gemini CLI login authentication
- Large responses chunked with RESPONSE_CHUNK_SIZE_KB (default 10 KB) and CACHE_TTL_MS (default 1 hour)
- PROJECT_ROOT path validation and DEBUG logging env toggles
- npm identifier gemini-researcher version 1.0.2
- default RESPONSE_CHUNK_SIZE_KB 10
What problem does it solve?
Primary coding agents burn context and cost when every deep research question is answered inline instead of through a dedicated research backend.
Who is it for?
Solo builders orchestrating Claude Code or Cursor who already use or can install Gemini CLI and want a research sidecar MCP.
Skip if: Teams forbidden from Gemini/Google APIs, or workflows that require fully inline single-model reasoning with no CLI dependency.
What do I get? / Deliverables
After stdio registration, your agent delegates research to Gemini CLI and receives chunked, cache-aware responses without holding full transcripts in the main session.
- Local stdio MCP research proxy connected to Gemini CLI
- Chunked research answers with configurable cache (default 3600000 ms TTL)
- Lower primary-agent context load during discovery and implementation research
Recommended MCP Servers
Journey fit
Research is the first obvious hook, but the server is a cross-phase offload pattern whenever you need external synthesis without growing primary context. Idea research is the canonical shelf for discovery queries; the npm stdio server is fundamentally agent-tooling that also serves validate and build lookups.
How it compares
Research-proxy MCP integration, not a bundled web-search skill inside the primary model.
Common Questions / FAQ
Who is io.github.capyBearista/gemini-researcher for?
Agent-first solo builders who want Gemini CLI to handle heavy research while the main coding agent stays lean and stateless.
When should I use io.github.capyBearista/gemini-researcher?
Use it whenever you need multi-step or large research during Idea discovery, validation, or build lookups without expanding the orchestrator context.
How do I add io.github.capyBearista/gemini-researcher to my agent?
npm install/run gemini-researcher as stdio MCP, set GEMINI_API_KEY or authenticate Gemini CLI, optionally set PROJECT_ROOT and chunk or cache env vars, then point Claude Code or Cursor at the server entry.