🏛️ Roundtable MCP (roundtable-mcp)

Role-Based Hybrid Multi-Model Orchestration MCP Server

roundtable-mcp is an open-source Model Context Protocol (MCP) server that orchestrates a panel of specialized AI models—combining local zero-cost models (via Ollama) with cloud APIs (Anthropic Claude, Google Gemini, OpenAI, DeepSeek, etc.).

Instead of voting on generic answers, Roundtable assigns distinct expert roles to each model (Architect, Implementation Engineer, Narrative Specialist, Systems Refactoring Expert, Deep Reasoning Specialist) and synthesizes their insights into unified advisory council reports.

---

✨ Key Features

• Role-Based Panel: Each model acts from a designated domain perspective rather than providing generic duplicate responses.
• Instant Model Onboarding: Onboard new models to ongoing tasks by querying targeted past note snippets (~400 tokens) from Obsidian memory instead of wasting 50,000+ tokens re-explaining conversation history.
• Hybrid Execution: Seamlessly mix local models (Ollama Gemma 4, Qwen Coder, Llama 3) with cloud APIs (Claude 3.7/4.6, Gemini 3.5, DeepSeek R1, GPT-4o).
• Parallel Fan-Out: run_roundtable queries all active panel models in parallel using Promise.allSettled() for fast execution and high fault tolerance.
• Config-Driven: Add or tweak models anytime by editing config.json without modifying code.
• Cross-Platform MCP: Native compatibility with Claude Code, Cursor, Windsurf, Antigravity, OpenCode, and VS Code.

---

📦 Quickstart

1. Installation

git clone https://github.com/YOUR_USERNAME/roundtable-mcp.git
cd roundtable-mcp
npm install
npm run build

2. Configuration

Copy config.example.json to config.json:

cp config.example.json config.json

API Key Options:

1. No API Keys Needed (100% Free Local Setup):

   • If using local Ollama models (provider: "ollama"), no keys or environment variables are required.

2. Direct in config.json:

   • You can put keys directly inside config.json using "api_key": "sk-...".

3. Via Environment Variables:

   • Or reference environment variables using "api_key_env": "ANTHROPIC_API_KEY".

export ANTHROPIC_API_KEY="sk-ant-..."
export DEEPSEEK_API_KEY="sk-..."
export OPENAI_API_KEY="sk-..."
export GEMINI_API_KEY="AIzaSy..."

---

🛠️ MCP Tool Reference

Tool Name	Description
run_roundtable	Runs a prompt across all enabled Roundtable experts in parallel and outputs a structured Advisory Council Report. Auto-logs to Obsidian if enabled.
run_cascade	Runs a sequential relay pipeline across expert models where each step builds upon the output of preceding steps (e.g. Developer -> QA -> CI/CD).
route_prompt	Routes prompt to a specific expert model or auto-detects based on keyword triggers ("gemini", "gemma", "claude", "deepseek", "gpt").
list_roundtable_models	Lists all registered expert models, their roles, and current provider status.
list_preset_roles	Lists built-in preset roles (QA Tester, CI/CD Engineer, Security Auditor, Database Architect, etc.).
search_obsidian_memory	Searches past Roundtable session notes in your Obsidian Vault for relevant context snippets.
save_to_obsidian	Manually or automatically logs custom notes & sessions directly into your Obsidian Vault.
roundtable_help	Displays an instant Quickstart Cheatsheet, command list, role catalog, and setup tips directly in chat.

---

⚡ Workflow Capabilities: Simultaneous & Cascading Workflows

1. 🔄 Simultaneous Multi-Tasking (Parallel)

You can designate distinct tasks to each expert role simultaneously in a single prompt:

"Architect: Design the schema. Coder: Write the GDScript implementation. QA: List edge-case test scenarios."

run_roundtable fans out the prompt to all assigned models concurrently, synthesizing their outputs in parallel.

2. 🌊 Cascading Sequential Relays (run_cascade)

In a cascading workflow, model tasks execute in sequential order where each role receives the prompt plus the cumulative outputs of all preceding roles:

1. Step 1 (Developer): Delivers initial code implementation.
2. Step 2 (QA Specialist): Evaluates Step 1's code and writes unit test suites against it.
3. Step 3 (CI/CD Engineer): Receives both the code and test suites to construct automated GitHub Actions pipelines.

# Example pipeline sequence
run_cascade(prompt: "Build an auth microservice", pipeline: ["architect", "qa_tester", "cicd_engineer"])

---

📋 Pre-Built Role Catalog (role_preset)

Instead of writing custom role descriptions, you can use built-in preset keys in config.json via "role_preset":

• qa_tester: QA & Edge-Case Testing Specialist
• cicd_engineer: DevOps & CI/CD Pipeline Specialist
• security_auditor: Cybersecurity & Vulnerability Auditor
• database_architect: Database Architect & Query Optimizer
• architect: Lead Systems Architect & Orchestrator
• performance_tuner: Low-Level Performance & Refactoring Specialist
• frontend_ux: Frontend & UI/UX Accessibility Specialist
• game_engineer: Game Engine & Mechanics Specialist
• tech_writer: Technical Documentation & API Writer

{
  "id": "claude-qa",
  "name": "Claude Sonnet 4.6",
  "role_preset": "qa_tester",
  "provider": "anthropic",
  "model": "claude-3-7-sonnet-20250219",
  "enabled": true
}

---

🎯 Assigning Specialized Skills to Models (skill)

Yes! You can assign specific domain skill guidelines, coding standards, or checklists to each model based on its role.

Use the "skill" (or "skill_prompt") property in config.json to inject customized instructions into that specific model's system prompt:

{
  "id": "gemma-godot",
  "name": "Gemma 4 (Local)",
  "role": "GDScript Implementation Specialist",
  "skill": "Follow Godot 4 GDScript standards: strict static typing, signal decoupling, and node path verification.",
  "provider": "ollama",
  "endpoint": "http://localhost:11434/api/generate",
  "model": "gemma4-26b-128k:latest",
  "enabled": true
}

---

📓 Obsidian Vault Integration (Long-Term Memory)

roundtable-mcp connects directly to your local Obsidian Vault to automatically log session notes and provide persistent memory across sessions—even for local models with smaller context windows!

Configuration in config.json:

{
  "obsidian": {
    "enabled": true,
    "vault_path": "C:/Users/yourname/Documents/ObsidianVault",
    "folder": "Roundtable/Sessions",
    "auto_save": true,
    "max_context_notes": 3
  }
}

• Auto-Log Sessions: Every run_roundtable session is formatted as a clean Markdown note complete with metadata, date, and tags (#roundtable #council #gemma #claude).
• Instant Model Onboarding: When introducing a new model or starting a new session, use search_obsidian_memory to fetch compact past note snippets (~400 tokens), onboarding the new expert instantly without spending 50,000+ tokens re-explaining chat history!
• Memory Retrieval: Keeps models with smaller context windows (e.g. 8k/16k local Ollama models) informed without context overflow or performance degradation.

🌐 Platform & Subscription Compatibility Guide

roundtable-mcp connects to models across different environments and subscriptions:

Platform / Tool	Native Subscriptions	Cross-Provider API Keys	Local Ollama
Google Antigravity	Gemini, Claude, GPT-OSS (Built-in)	❌ Not needed inside Antigravity	✅ Via Ollama MCP
Claude Code CLI	Claude (Covered by plan)	🔑 Requires API keys or free GitHub/AI Studio tokens for GPT-4o / Gemini	✅ Free via roundtable-mcp
OpenAI Codex CLI	GPT-4o (Covered by plan)	🔑 Requires API keys or free AI Studio tokens for Claude / Gemini	✅ Free via roundtable-mcp
Cursor / Windsurf	Claude, GPT-4o, Gemini (Bundled)	🔑 Optional for custom endpoints	✅ Free via roundtable-mcp

• Single-Provider CLI Tools (Claude Code / Codex): Your subscription covers the primary model (Claude or GPT). To query third-party models, configure keys in config.json (or use 100% free options like GitHub Models GITHUB_TOKEN, Google AI Studio GEMINI_API_KEY, or local Ollama).
• Multi-Model Native Agents (Antigravity): Gemini, Claude, and GPT-OSS are provided directly by the platform without separate API keys.

---

🔧 Integrating with AI Coding Platforms

Claude Code / Claude Desktop / Antigravity / OpenCode

Add roundtable-mcp to your MCP configuration file (mcpServers section):

{
  "mcpServers": {
    "roundtable": {
      "command": "node",
      "args": ["/path/to/roundtable-mcp/dist/index.js"],
      "env": {
        "ANTHROPIC_API_KEY": "sk-ant-...",
        "DEEPSEEK_API_KEY": "sk-...",
        "OPENAI_API_KEY": "sk-...",
        "GEMINI_API_KEY": "AIzaSy..."
      }
    }
  }
}

---

⚙️ Adding New Models to config.json

You can add any OpenAI-compatible provider (DeepSeek, Groq, Together AI, Mistral, LM Studio, vLLM) easily:

{
  "id": "qwen-coder-local",
  "name": "Qwen 2.5 Coder (Local)",
  "role": "Low-Level Shader & GPU Optimization Specialist",
  "provider": "ollama",
  "endpoint": "http://localhost:11434/api/generate",
  "model": "qwen2.5-coder:32b",
  "enabled": true
}

---

💡 100% Free Multi-Model Stack Example

You can run a complete, multi-model Roundtable setup entirely for free by combining local Ollama models with free cloud API tiers (GitHub Models, Google AI Studio, Groq):

{
  "models": [
    {
      "id": "gpt4o-free",
      "name": "GPT-4o (GitHub Free)",
      "role": "Systems Performance & Refactoring Specialist",
      "provider": "openai-compatible",
      "endpoint": "https://models.github.ai/inference/chat/completions",
      "model": "gpt-4o",
      "api_key_env": "GITHUB_TOKEN",
      "enabled": true
    },
    {
      "id": "gemini-free",
      "name": "Gemini 2.5 Flash (Google Free)",
      "role": "Lead Architect & Synthesizer",
      "provider": "google",
      "model": "gemini-2.5-flash",
      "api_key_env": "GEMINI_API_KEY",
      "enabled": true
    },
    {
      "id": "gemma-local",
      "name": "Gemma 4 (Local Ollama)",
      "role": "Primary Implementation Engineer",
      "provider": "ollama",
      "endpoint": "http://localhost:11434/api/generate",
      "model": "gemma4-26b-128k:latest",
      "enabled": true
    },
    {
      "id": "deepseek-free",
      "name": "DeepSeek R1 (Groq Free)",
      "role": "Deep Reasoning & Math Specialist",
      "provider": "openai-compatible",
      "endpoint": "https://api.groq.com/openai/v1/chat/completions",
      "model": "deepseek-r1-distill-llama-70b",
      "api_key_env": "GROQ_API_KEY",
      "enabled": true
    }
  ]
}

---

📄 License

MIT License. Feel free to use, modify, and distribute!