mcp-consultant

Simplified fork of PAL MCP Server for uctoteka project. Supports only Gemini and OpenAI providers with predefined model configurations.

Features

• Gemini CLI integration: gemini-3-pro-preview (default), gemini-3-flash-preview (fast)
• OpenAI/Codex integration: gpt-5.2 with xhigh thinking mode
• Removed providers: Azure OpenAI, XAI, DIAL, OpenRouter, Custom API
• Fixed model mode: No "auto" mode - models are explicitly configured
• MCP protocol: Full Model Context Protocol support for Claude Code

Installation

cd /home/pavel/vyvoj_sw/mcp-consultant
pip install --break-system-packages -e .

Configuration

CLI Clients

Located in conf/cli_clients/:

Gemini (gemini.json):

{
  "name": "gemini",
  "command": "gemini",
  "roles": {
    "default": {"role_args": ["--model", "gemini-3-pro-preview"]},
    "flash": {"role_args": ["--model", "gemini-3-flash-preview"]}
  }
}

Codex (codex.json):

{
  "name": "codex",
  "command": "codex",
  "additional_args": ["exec", "--json", "--model", "gpt-5.2"],
  "roles": {
    "default": {"role_args": ["--config", "model_reasoning_effort=\"xhigh\""]}
  }
}

Environment Variables

# Required for actual API calls (optional for listmodels/version)
export GEMINI_API_KEY="your_key_here"
export OPENAI_API_KEY="your_openai_key_here"

Usage

Standalone

mcp-consultant

Via Claude Code

The server is registered in uctoteka-clink-consult plugin at:

/home/pavel/vyvoj_sw/uctoteka_app/claude-marketplace/uctoteka-clink-consult/mcp/pal.mcp.json

Configuration:

{
  "mcpServers": {
    "pal": {
      "command": "/home/pavel/.local/bin/mcp-consultant",
      "env": {
        "CLI_CLIENTS_CONFIG_PATH": "${CLAUDE_PLUGIN_ROOT}/cli_clients",
        "DEFAULT_MODEL": "gemini-3-pro-preview"
      }
    }
  }
}

Available Tools

• clink - Bridge to external AI CLIs (Gemini, Codex)
• listmodels - List available models per provider
• version - Show server version and configuration

Model Comparison

Model	Thinking	Latence	Use case
gemini-3-pro-preview	high	~47s	Complex analysis
gemini-3-flash-preview	-	~13s	Quick queries
gpt-5.2 (codex)	xhigh	~31s	Max reasoning, code + web

Development

Project structure:

mcp-consultant/
├── clink/           # CLI integration (agents, parsers, registry)
├── providers/       # Model providers (Gemini, OpenAI)
├── tools/           # MCP tools (clink, listmodels, version)
├── conf/            # Configuration files
├── systemprompts/   # System prompts for CLIs
├── server.py        # Main MCP server
├── config.py        # Configuration constants
└── pyproject.toml   # Package definition

Original project: PAL MCP Server

Fork changes:

• Removed Azure OpenAI, XAI, DIAL, OpenRouter, Custom providers
• Fixed DEFAULT_MODEL to gemini-3-pro-preview
• Disabled auto mode (IS_AUTO_MODE = False)
• Server starts without API keys (for listmodels/version)
• Simplified to 2 CLI clients: gemini, codex

Version

1.0.0 - Initial fork for uctoteka