Gemini CLI

Building AI Scientists with Gemini CLI and ToolUniverse

Overview

Gemini CLI Integration enables powerful command-line-based scientific research through the Model Context Protocol (MCP). This approach provides a terminal-based interface for scientific research while leveraging Gemini’s advanced reasoning capabilities and ToolUniverse’s comprehensive scientific tools ecosystem.

┌─────────────────┐
│   Gemini CLI    │ ← Command Line Interface & Reasoning
│                 │
└─────────┬───────┘
          │ MCP Protocol
          │
┌─────────▼───────┐
│ ToolUniverse     │ ← MCP Server
│   MCP Server     │
└─────────┬───────┘
          │
┌─────────▼───────┐
│ 600+ Scientific │ ← Scientific Tools Ecosystem
│     Tools       │
└─────────────────┘

Benefits of Gemini CLI Integration:

  • Command-Line Efficiency: Fast, scriptable scientific research workflows

  • Advanced Reasoning: Gemini’s sophisticated reasoning for complex scientific problems

  • Comprehensive Tools: Access to 600+ scientific tools across multiple domains

  • Automated Execution: Direct tool execution through natural language commands

  • Batch Processing: Handle multiple research tasks efficiently

Prerequisites

Before setting up Gemini CLI integration, ensure you have:

  • Gemini CLI: Installed and running on your system

  • Python 3.10+: Required for ToolUniverse

  • UV Package Manager: For dependency management

  • System Requirements: macOS, Windows, or Linux

  • API Keys: Valid API keys for external services (if required by specific tools)

Installation and Setup

Step 1: Install Gemini CLI

Install Gemini CLI following the official installation Tutorial:

# Follow the installation Tutorial at:
# https://github.com/google-gemini/gemini-cli

Verify installation:

gemini --version

Step 2: Install UV and ToolUniverse

# Install UV package manager
curl -LsSf https://astral.sh/uv/install.sh | sh

# Create working directory for ToolUniverse
# uv will automatically create and manage a virtual environment (.venv) inside this directory
mkdir -p /path/to/tooluniverse-env
uv --directory /path/to/tooluniverse-env pip install tooluniverse

# Verify installation
uv --directory /path/to/tooluniverse-env run python -c "import tooluniverse; print('ToolUniverse installed successfully')"

Step 3: Locate Gemini CLI Configuration

The Gemini CLI uses configuration files to manage MCP servers. You can configure them:

  • Globally: In the ~/.gemini/settings.json file

  • Per project: In your project’s .gemini/settings.json file

Create or open your settings.json file:

# Global configuration
mkdir -p ~/.gemini
touch ~/.gemini/settings.json

# Or project-specific configuration
mkdir -p .gemini
touch .gemini/settings.json

Step 4: Configure ToolUniverse MCP Server

Add the ToolUniverse MCP server configuration to your settings.json:

Basic Configuration:

{
  "mcpServers": {
    "tooluniverse": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/tooluniverse-env",  # Working directory: uv manages .venv here
        "run",
        "tooluniverse-smcp-stdio"
      ]
    }
  }
}

Optimized Configuration for Research Users (Recommended):

{
  "mcpServers": {
    "tooluniverse": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/tooluniverse-env",  # Working directory: uv manages .venv here
        "run",
        "tooluniverse-smcp-stdio",
        "--exclude-tool-types",
        "PackageTool",
        "--hook-type",
        "SummarizationHook"
      ],
      "env": {
        "AZURE_OPENAI_API_KEY": "your-azure-openai-api-key",
        "AZURE_OPENAI_ENDPOINT": "https://your-resource.openai.azure.com"
      }
    }
  }
}

Configuration Benefits:

  • --exclude-tool-types PackageTool: Removes package management tools to save context window space if you don’t have coding needs

  • --hook-type SummarizationHook: Provides summary of the output that is too long to fit in the context window

  • AZURE_OPENAI_API_KEY and AZURE_OPENAI_ENDPOINT: Required for SummarizationHook functionality

Important: Gemini CLI 500 Tool Limit

Gemini CLI has a 500 tool limit per MCP server. ToolUniverse provides 750+ tools by default, which exceeds this limit. You must use one of the following solutions to work within this limit:

Solution 1: Space Configuration (Recommended for Gemini CLI)

Use the pre-configured gemini-essential.yaml Space to access ~400-450 essential scientific tools.

Download the configuration file:

Download gemini-essential.yaml from the ToolUniverse repository and save it to your project directory (e.g., ./examples/spaces/gemini-essential.yaml).

Alternatively, if you have cloned the ToolUniverse repository, the file is already available at examples/spaces/gemini-essential.yaml.

Configuration:

{
  "mcpServers": {
    "tooluniverse": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/tooluniverse-env",
        "run",
        "tooluniverse-smcp-stdio",
        "--load",
        "./examples/spaces/gemini-essential.yaml"
      ]
    }
  }
}

Benefits of gemini-essential.yaml:

  • Under 500 tools: Carefully selected ~400-450 essential tools, well within Gemini CLI’s limit

  • Ready to use: No need to manually configure tool selection

  • Direct access: All selected tools are directly available without additional discovery steps

What’s Included:

  • Core scientific databases (FDA, OpenTargets, UniProt, PubChem)

  • Literature search tools (PubMed, Europe PMC, OpenAlex, Semantic Scholar)

  • Clinical data tools (ClinicalTrials.gov, FAERS)

  • Structural biology tools

  • Genomics and molecular biology tools

Solution 2: Compact Mode (Alternative for Minimal Context Usage)

For maximum context window efficiency, use compact mode which exposes only 4-5 core tools:

{
  "mcpServers": {
    "tooluniverse": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/tooluniverse-env",
        "run",
        "tooluniverse-smcp-stdio",
        "--compact-mode"
      ]
    }
  }
}

Compact Mode Benefits:

  • 99% reduction: Only 4-5 tools exposed instead of 750+

  • Full functionality: All 750+ tools still accessible via execute_tool

  • Minimal context usage: Ideal for AI agents with limited context windows

  • Progressive disclosure: Use list_tools, grep_tools, and get_tool_info to discover tools on demand

  • Well within limit: 4-5 tools is far below the 500 tool limit

Compact Mode Tools:

  • list_tools: List all available tools

  • grep_tools: Search for tools by name or pattern

  • get_tool_info: Get detailed information about specific tools

  • execute_tool: Execute any ToolUniverse tool

See also

For detailed information about compact mode, see Compact Mode Guide

Which Solution Should You Choose?

Feature

Space Configuration

Compact Mode

Tools Exposed

~400-450 tools

4-5 tools

Direct Access

Yes, all selected tools

Via execute_tool

Context Usage

Medium

Minimal (99% reduction)

Tool Discovery

Not needed

Required (list_tools, grep_tools)

Best For

Users who want direct access to common tools

Users who want minimal context usage

Recommendation:

  • Use Space Configuration if you want direct access to commonly used scientific tools

  • Use Compact Mode if you want minimal context usage and don’t mind using tool discovery commands

Important Configuration Notes:

  • Replace /path/to/tooluniverse-env with your actual ToolUniverse working directory

  • The working directory is where uv will create and manage the virtual environment (.venv)

  • The path should point to the directory where your Gemini CLI environment is set up

  • Use absolute paths for better reliability

Example Configuration:

{
  "mcpServers": {
    "tooluniverse": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/tooluniverse-env",  # Working directory
        "run",
        "tooluniverse-smcp-stdio"
      ]
    }
  }
}

Step 5: Configure Context File (GEMINI.md)

Create a context file to provide project-specific instructions for scientific research:

  1. Create GEMINI.md file in your project root: .. code-block:: bash

    vim GEMINI.md

  2. Add ToolUniverse-specific context: .. code-block:: markdown

# ToolUniverse Scientific Research Project

## General Instructions

  • Leverage ToolUniverse’s scientific tools ecosystem for evidence-based research

  • Use tools from tooluniverse mcp server first

  • Cross-validate findings across multiple tools and databases

  • Use appropriate scientific terminology

  1. Verify context loading: .. code-block:: bash

    /memory show

This will display the loaded context files and their content.

Step 6: Start Gemini CLI

After saving the configuration:

  1. Start Gemini CLI: .. code-block:: bash

    gemini

  2. Verify MCP Server Connection: .. code-block:: bash

    /mcp

This will display: - Server connection status (CONNECTED or DISCONNECTED) - Available ToolUniverse tools - Configuration details - Any connection errors

Step 7: Verify Integration

Test the integration by asking Gemini to:

  1. List available tools: .. code-block:

    What scientific tools are available?

  2. Execute a simple tool: .. code-block:

    Search for information about Alzheimer's disease

  3. Perform complex research: .. code-block:

    Find recent papers about CRISPR gene editing in cancer therapy

Settings and Configuration

Tool Selection Strategies

Optimize tool usage for better performance:

Selective Tool Loading: - Load only relevant tools for specific research domains - Reduce context window usage - Improve response times

Gemini CLI 500 Tool Limit: - See the gemini-500-tool-limit section in Step 4 above for detailed solutions - Use gemini-essential.yaml Space configuration or compact mode to stay within limits

Example Tool Selection:

{
  "mcpServers": {
    "tooluniverse": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/tooluniverse-env",  # Working directory: uv manages .venv here
        "run",
        "tooluniverse-smcp-stdio"
      ],
      "includeTools": [
        "EuropePMC_search_articles",
        "ChEMBL_search_similar_molecules",
        "openalex_literature_search",
        "search_clinical_trials"
      ]
    }
  }
}

Custom Tool Sets

Create custom tool sets for specific research domains:

{
  "mcpServers": {
    "tooluniverse-literature": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/tooluniverse-env",  # Working directory: uv manages .venv here
        "run",
        "tooluniverse-smcp-stdio"
      ],
      "includeTools": [
        "EuropePMC_search_articles",
        "openalex_literature_search",
        "PubTator3_LiteratureSearch"
      ]
    },
    "tooluniverse-compounds": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/tooluniverse-env",  # Working directory: uv manages .venv here
        "run",
        "tooluniverse-smcp-stdio"
      ],
      "includeTools": [
        "ChEMBL_search_similar_molecules",
        "FDA_get_drug_names_by_indication",
        "drugbank_search"
      ]
    }
  }
}

Multiple MCP Servers

Run multiple ToolUniverse instances for different purposes:

{
  "mcpServers": {
    "tooluniverse-research": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/tooluniverse-env",  # Working directory: uv manages .venv here
        "run",
        "tooluniverse-smcp-stdio"
      ],
      "timeout": 30000
    },
    "tooluniverse-analysis": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/tooluniverse-env",  # Working directory: uv manages .venv here
        "run",
        "tooluniverse-smcp-stdio"
      ],
      "timeout": 45000
    }
  }
}

Troubleshooting

Common Issues and Solutions

MCP Server Not Loading: - Verify ToolUniverse installation path - Check UV package manager installation - Ensure proper JSON syntax in configuration - Check Gemini CLI logs for errors

No Tools Discovered: - Verify the ToolUniverse MCP server is working - Check if your includeTools or excludeTools filter is too restrictive - Ensure all ToolUniverse dependencies are installed - Review your ToolUniverse installation and configuration

Tools Not Executing: - Check your env configuration for API keys - Verify network connectivity to the required scientific APIs - Increase the timeout value in your configuration - Review parameter validation errors in the CLI output

Performance Issues: - Increase the timeout value in the configuration - Use includeTools to load only the necessary tools - Check network connectivity and latency to external APIs - Be mindful of potential rate limits for external services

Debug Mode

Run the Gemini CLI with the –debug flag for detailed information:

gemini --debug

This provides verbose output about: - MCP server connection attempts - The tool discovery process - Tool execution details and errors

Tips

Tool Selection: Use includeTools to load only the tools you need for a specific task.

Status Check: Use /mcp to monitor server status and available tools.

Debug Mode: Use the –debug flag for verbose troubleshooting information.

Tip

Start with simple queries: Begin with basic tool discovery and simple research questions to understand the integration, then progress to complex multi-step workflows as you become more familiar with the capabilities.

Note

Command-Line Efficiency: Gemini CLI provides a powerful command-line interface for scientific research, enabling both interactive exploration and automated batch processing of research tasks.

Scientific Research Capabilities

Drug Discovery and Development

Gemini CLI with ToolUniverse enables comprehensive drug discovery workflows:

Target Identification: - Disease analysis and EFO ID lookup - Target discovery and validation - Literature-based target assessment

Drug Analysis: - Drug information retrieval from multiple databases - Safety profile analysis - Drug interaction checking - Clinical trial data access

Example Workflow:

User: Find FDA-approved drugs that target the EGFR protein and show me their chemical structures

[Gemini uses ToolUniverse tools to:]
1. Search FDA database for EGFR-targeting drugs
2. Retrieve drug information and structures
3. Analyze chemical properties
4. Provide comprehensive drug profiles

Genomics and Molecular Biology

Access comprehensive genomics tools for molecular research:

Gene Analysis: - Gene information from UniProt - Protein structure analysis - Expression pattern analysis - Pathway involvement

Molecular Interactions: - Protein-protein interactions - Drug-target interactions - Pathway analysis - Functional annotation

Example Workflow:

User: Analyze the BRCA1 gene and its role in cancer development

[Gemini uses ToolUniverse tools to:]
1. Get BRCA1 gene information from UniProt
2. Analyze protein structure and function
3. Find protein interactions
4. Identify cancer-related pathways
5. Search for therapeutic targets

Literature Research and Analysis

Comprehensive literature search and analysis capabilities:

Literature Search: - PubMed searches - Semantic Scholar integration - Europe PMC access - Citation analysis

Content Analysis: - Abstract summarization - Key finding extraction - Trend analysis - Gap identification

Example Workflow:

User: Find recent papers about CRISPR gene editing in cancer therapy published in the last 2 years

[Gemini uses ToolUniverse tools to:]
1. Search PubMed for recent papers
2. Filter by publication date
3. Analyze abstracts and key findings
4. Identify research trends
5. Provide comprehensive review

Clinical Research and Trials

Access clinical trial data and regulatory information:

Clinical Trials: - ClinicalTrials.gov searches - Trial status and results - Patient population analysis - Outcome assessment

Regulatory Information: - FDA drug approvals - Safety warnings - Labeling information - Adverse event reports

Example Workflow:

User: Find ongoing clinical trials for Alzheimer's disease treatments in the United States

[Gemini uses ToolUniverse tools to:]
1. Search ClinicalTrials.gov
2. Filter by Alzheimer's disease and US location
3. Analyze trial designs and outcomes
4. Check FDA approvals
5. Assess safety profiles

Multi-Step Research Workflows

Gemini CLI excels at complex, multi-step research workflows:

Hypothesis-Driven Research: 1. Formulate research hypothesis 2. Design experimental approach 3. Gather supporting evidence 4. Validate findings 5. Generate conclusions

Comparative Analysis: 1. Identify comparison targets 2. Gather data for each target 3. Perform comparative analysis 4. Identify differences and similarities 5. Draw conclusions

Example Multi-Step Workflow:

User: I'm researching potential drug targets for Parkinson's disease. Can you:
1. Find recent papers on Parkinson's disease drug targets
2. Search for compounds that interact with those targets
3. Check if there are any ongoing clinical trials for those compounds

[Gemini executes multi-step workflow:]
1. Search literature for Parkinson's disease drug targets
2. Identify key targets and mechanisms
3. Search for compounds interacting with targets
4. Check clinical trial status
5. Provide comprehensive analysis

Iterative Research and Refinement

Gemini can iteratively refine research based on intermediate results:

Adaptive Research: - Start with broad search - Refine based on initial results - Focus on promising directions - Validate findings with additional tools

Example Iterative Workflow:

User: Investigate potential treatments for rare genetic diseases

[Gemini performs iterative research:]
1. Initial broad search for rare disease treatments
2. Analyze results and identify patterns
3. Focus on gene therapy approaches
4. Investigate specific gene therapy trials
5. Analyze safety and efficacy data

Batch Processing and Automation

Gemini CLI enables batch processing of multiple research tasks:

Automated Workflows: - Process multiple queries simultaneously - Generate comprehensive reports - Export results in various formats - Schedule recurring research tasks

Example Batch Processing:

User: Analyze the following genes for cancer association: BRCA1, BRCA2, TP53, EGFR, KRAS

[Gemini processes batch analysis:]
1. Analyze each gene individually
2. Compare cancer associations
3. Identify common pathways
4. Generate comparative report