Gemini CLI vs Claude Code Feature Parity Research

Core Idea

Investigate and document how closely Gemini CLI capabilities align with Claude Code’s extensibility ecosystem, specifically examining:

• Skills - Custom capability packages
• Commands - Slash command definitions
• MCP (Model Context Protocol) - Tool integration layer
• Templates - Reusable content scaffolds
• Scripts - Automation helpers

The goal is to create portable, cross-compatible tooling that could be distributed through a plugin marketplace.

Why This Matters

1. Reduce vendor lock-in - Skills/commands that work across AI CLI tools provide flexibility
2. Maximize investment - Time spent building Claude Code extensions could benefit Gemini users too
3. Community benefit - A unified plugin marketplace could accelerate AI-assisted development adoption
4. Future-proofing - As AI CLI tools evolve, understanding common patterns helps adaptation

Research Questions

• Does Gemini CLI have an equivalent to Claude Code’s ~/.claude/commands/ structure?
• How does Gemini handle MCP server integration?
• What’s Gemini CLI’s approach to skills/plugins?
• Are there template or scaffolding mechanisms in Gemini CLI?
• What scripting hooks exist in Gemini CLI?

Feature Comparison Matrix (To Research)

Potential Marketplace Concept

If feature parity exists, a unified marketplace could offer:

• Cross-platform skill packages
• Adapter layers for tool-specific differences
• Shared MCP server configurations
• Universal template libraries

Related Concepts

• Claude Code Skills Architecture
• Gemini CLI Extensions
• MCP Protocol Specification
• AI Tool Interoperability
• Plugin Marketplace Design (npm, VS Code extensions model)

Next Steps

1. Deep-dive into Gemini CLI documentation for extension capabilities
2. Compare MCP support between tools
3. Prototype a simple cross-compatible command
4. Research existing marketplace models (Raycast, VS Code, npm)
5. Draft specification for portable AI CLI plugin format

Portability Test: obsidian-vault-manager-plugin

Testing whether the obsidian-vault-manager-plugin can be ported to Gemini CLI.

Plugin Components to Test

Action Items

• Test 1: Gemini CLI Command Structure
  • Check if Gemini CLI has a commands directory similar to .claude/commands/
  • Test if markdown-based command definitions work
  • Document: gemini --help and configuration paths
• Test 2: MCP Support in Gemini CLI
  • Verify if Gemini CLI supports MCP protocol
  • Check Docker MCP Toolkit compatibility with Gemini
  • Test: Can Gemini access mcp__MCP_DOCKER__obsidian_* tools?
• Test 3: Port Simple Command
  • Start with /idea command (simplest, no external dependencies)
  • Create Gemini-compatible version
  • Compare: execution flow, output format, success rate
• Test 4: Port YouTube Capture
  • Test /youtube-note which uses MCP Docker YouTube tools
  • Dependencies: mcp__MCP_DOCKER__get_video_info, get_transcript
  • Check if same MCP tools accessible from Gemini
• Test 5: Skills/Plugins Architecture
  • Research Gemini CLI extension/plugin system
  • Check for SKILL.md equivalent
  • Test if Gemini has tool allowlists like Claude Code

Potential Compatibility Layers

┌─────────────────────────────────────┐
│     Universal Plugin Package        │
├─────────────────────────────────────┤
│  commands/     (markdown files)     │
│  scripts/      (bash - universal)   │
│  templates/    (text - universal)   │
├─────────────────────────────────────┤
│         Adapter Layer               │
├──────────────┬──────────────────────┤
│ Claude Code  │    Gemini CLI        │
│ Adapter      │    Adapter           │
└──────────────┴──────────────────────┘

Key Questions to Answer

1. Does Gemini CLI have a .gemini/ config directory?
2. Can Gemini CLI source bash scripts from a plugin?
3. Does Gemini support MCP servers (especially Docker MCP)?
4. How does Gemini handle tool permissions/allowlists?
5. Is there a Gemini equivalent to marketplace.json?

Research Commands to Run

# Check Gemini CLI config locations
gemini --version
ls -la ~/.gemini/ 2>/dev/null || echo "No .gemini directory"
gemini config list

# Check MCP support
gemini mcp list  # If MCP supported
gemini tools list  # List available tools

# Test simple prompt with tool use
gemini -p "What tools do you have access to?"

DoubleCopy Multi-Agent Testing Plan

DoubleCopy v1.0.34 currently uses Claude Code CLI exclusively. The roadmap includes multi-agent support for Gemini CLI, OpenAI Codex, and Qwen Code. This section documents the testing plan to validate Gemini CLI compatibility.

Current DoubleCopy Architecture

DoubleCopy.app
    ↓
capture.sh (generated)
    ↓
claude -p "/capture $CONTENT"    ← Agent-specific
    ↓
/capture command → MCP tools → Obsidian note

Key Discovery: DoubleCopy uses slash command invocation (claude -p "/capture $CONTENT") which properly follows templates. For Gemini CLI to work, it needs:

1. Non-interactive prompt mode (gemini -p "..." or equivalent)
2. Access to same MCP tools (obsidian, youtube-transcript, etc.)
3. Either slash command support OR direct prompt execution

Gemini CLI Compatibility Tests

Test 1: Basic Non-Interactive Execution

Objective: Verify Gemini CLI can execute prompts non-interactively

# Test 1a: Check if Gemini has non-interactive mode
gemini --help | grep -i "prompt\|non-interactive\|-p"

# Test 1b: Test simple non-interactive execution
gemini -p "Hello, respond with just 'OK'" 2>&1

# Test 1c: Test with longer prompt
gemini -p "Create a simple markdown note with title 'Test' and one bullet point"

Expected Result: Gemini should execute and return output without interactive shell

Status: ⬜ Not tested

Test 2: MCP Server Access

Objective: Verify Gemini CLI can access Docker MCP servers

# Test 2a: Check if Gemini supports MCP
gemini mcp list 2>&1 || echo "MCP not supported or different command"

# Test 2b: List available tools
gemini tools list 2>&1

# Test 2c: Test MCP tool execution
gemini -p "Use the obsidian tool to list files in the vault root directory"

Key Tools Needed for DoubleCopy:

• mcp__MCP_DOCKER__obsidian_append_content - Create notes
• mcp__MCP_DOCKER__get_video_info - YouTube metadata
• mcp__MCP_DOCKER__get_transcript - YouTube transcript
• mcp__MCP_DOCKER__firecrawl_scrape - Web article fetching

Status: ⬜ Not tested

Test 3: Simple Capture Test (Idea Note)

Objective: Test Gemini creating a simple Obsidian note

# Test 3a: Direct prompt capture (no slash command)
CONTENT="Test idea for Gemini CLI integration"
DATE=$(date "+%Y-%m-%d")

gemini -p "Create a markdown note in Obsidian vault with:
- Filename: ${DATE}-gemini-test-idea.md
- Frontmatter with tags: idea, AI, testing
- Title: 'Gemini CLI Test Idea'
- Content: ${CONTENT}
Use the obsidian_append_content MCP tool to create the file."

# Test 3b: Verify note created
ls -la ~/Documents/Obsidian/Claudecode/${DATE}-gemini-*.md

Status: ⬜ Not tested

Test 4: YouTube Capture Test

Objective: Test Gemini capturing a YouTube video with transcript

# Test 4a: YouTube capture with MCP tools
VIDEO_URL="https://www.youtube.com/watch?v=dQw4w9WgXcQ"

gemini -p "Capture this YouTube video to Obsidian:
URL: ${VIDEO_URL}

1. Use get_video_info to get title, channel, duration
2. Use get_transcript to get the transcript
3. Create a structured note with:
   - Thumbnail image
   - Video metadata
   - Key points from transcript
4. Use obsidian_append_content to save the note"

# Test 4b: Check if note was created with thumbnail
grep -l "maxresdefault.jpg" ~/Documents/Obsidian/Claudecode/*.md | tail -1

Status: ⬜ Not tested

Test 5: capture.sh Modification for Multi-Agent

Objective: Design capture.sh to support agent switching

Current capture.sh (Claude only):

claude -p "/capture $CONTENT"

Proposed Multi-Agent capture.sh:

#!/bin/bash
CONTENT="$1"
AGENT="${DOUBLECOPY_AGENT:-claude}"  # Default to claude

case "$AGENT" in
    claude)
        # Slash command invocation (works with Claude)
        claude -p "/capture $CONTENT"
        ;;
    gemini)
        # Direct prompt (if Gemini doesn't support slash commands)
        # Load capture prompt from template
        PROMPT=$(cat ~/.doublecopy/prompts/capture-gemini.md)
        PROMPT="${PROMPT//\$CONTENT/$CONTENT}"
        gemini -p "$PROMPT"
        ;;
    *)
        echo "Unknown agent: $AGENT"
        exit 1
        ;;
esac

Action Items:

• Create ~/.doublecopy/prompts/capture-gemini.md template
• Test agent switching via environment variable
• Compare output quality between Claude and Gemini

Status: ⬜ Not implemented

Test 6: DoubleCopy App Modification

Objective: Add Gemini CLI support to DoubleCopy.app

Swift Changes Required (in main.swift):

// 1. Add Gemini to CLIAgent enum
enum CLIAgent: String, CaseIterable {
    case claude = "Claude Code"
    case gemini = "Gemini CLI"
}

// 2. Add Gemini check function
func checkGeminiCLI() -> Bool {
    let result = shell("which gemini || ls ~/.gemini/bin/gemini 2>/dev/null")
    return !result.isEmpty
}

// 3. Modify capture.sh generation
func installCaptureScript() {
    let agent = selectedAgent // From preferences
    let captureCommand: String

    switch agent {
    case .claude:
        captureCommand = "claude -p \"/capture $CONTENT\""
    case .gemini:
        captureCommand = "gemini -p \"$(cat ~/.doublecopy/prompts/capture.md)\""
    }
    // ... rest of script generation
}

Status: ⬜ Not implemented

Testing Checklist Summary

Success Criteria

For Gemini CLI to be considered “DoubleCopy Compatible”:

1. ✅ Can execute prompts non-interactively
2. ✅ Can access MCP Docker tools (obsidian, youtube, etc.)
3. ✅ Can create properly formatted Obsidian notes
4. ✅ Output quality comparable to Claude Code
5. ✅ Response time acceptable (<30 seconds for simple capture)

Fallback Strategy

If Gemini CLI lacks certain capabilities:

Captured: 2025-12-30 Updated: 2025-12-30 - Added obsidian-vault-manager portability test plan Updated: 2025-12-30 - Added DoubleCopy multi-agent testing plan with Gemini CLI action items