PAL MCP

A Provider Abstraction Layer (PAL) MCP server that orchestrates multiple AI models (Gemini, OpenAI, Azure, Grok, Ollama, OpenRouter, DIAL) within a single context. Enables multi-model collaboration, conversation continuity across AI providers, and workflow automation for code analysis, debugging, planning, and reviews.

Key Features

• Multi-Model Orchestration - Coordinate Gemini Pro, GPT-5, O3, and 50+ other models in single conversations
• Conversation Continuity - Full context flows across tools and models, preserving thread history
• CLI-to-CLI Bridge (clink) - Connect external AI CLIs (Gemini CLI, Codex CLI, Claude Code) as subagents
• Context Revival - Continue conversations even after context resets by having other models "remind" the primary AI
• Extended Context Windows - Leverage Gemini (1M tokens) or O3 (200K tokens) for massive codebases
• Vision Support - Analyze screenshots, diagrams, and visual content with vision-enabled models
• Local Model Support - Run Llama, Mistral, or other models via Ollama for complete privacy

Core Tools

Collaboration & Planning (enabled by default)

• clink - Bridge to external AI CLIs, spawn subagents with role specialization
• chat - Multi-turn brainstorming and second opinions with any model
• thinkdeep - Extended reasoning with configurable thinking modes
• planner - Break down complex projects into structured plans
• consensus - Multi-model debate with stance steering

Code Quality (enabled by default)

• codereview - Professional multi-pass reviews with severity levels
• precommit - Validate changes before committing
• debug - Systematic root cause analysis with confidence tracking

Development Tools (disabled by default)

• analyze - Architecture and dependency analysis
• refactor - Intelligent code refactoring
• testgen - Comprehensive test generation
• secaudit - Security audits with OWASP Top 10
• docgen - Documentation generation

Utilities

• apilookup - Force current-year API/SDK documentation lookups
• challenge - Critical analysis mode to prevent reflexive agreement
• tracer - Static analysis for call-flow mapping

Example Workflows

Multi-model code review:

Perform a codereview using gemini pro and o3, then use planner to create a fix strategy

Collaborative debugging:

Debug this race condition with max thinking mode, then validate the fix with precommit

CLI subagent for isolated tasks:

clink with codex codereviewer to audit auth module for security issues

Consensus-driven decisions:

Use consensus with gpt-5 and gemini-pro to decide: dark mode or offline support next

Configuration

Tools can be enabled/disabled via DISABLED_TOOLS environment variable. Default disabled: analyze,refactor,testgen,secaudit,docgen,tracer. Set to empty string to enable all tools.

Thinking modes: low, medium, high, max - control reasoning depth vs cost.