Skip to main content
Glama

MCP Ollama Consult Server

by Atomic-Germ
OverviewInspectSchemaRelated ServersScoreDiscussions
Hybrid
MIT License

MCP Ollama Consult Server

An intelligent MCP server for consulting with Ollama models and enabling multi-perspective AI reasoning

License: MIT TypeScript MCP CI/CD

Overview

MCP Ollama Consult is a Model Context Protocol (MCP) server that enables AI agents to consult with multiple Ollama models for diverse perspectives, reasoning chains, and collaborative problem-solving. It provides powerful tools for sequential consultation workflows and persistent memory management.

Key Features

  • šŸ¤ Multi-Model Consultation - Consult with any available Ollama model

  • šŸ“Š Model Comparison - Run identical prompts against multiple models simultaneously

  • šŸ§  Sequential Reasoning Chains - Execute complex multi-step reasoning workflows

  • šŸ’¾ Persistent Memory - Store and retrieve consultation results across sessions

  • šŸ”— Flexible Integration - Works with any MCP-compatible client or framework

  • āš” Timeout Management - Configurable timeouts for complex reasoning tasks

  • šŸŽÆ Demo Client - Built-in demo for testing and exploration

Installation

# Clone the repository git clone https://github.com/Atomic-Germ/mcp-consult.git cd mcp-consult # Install dependencies npm install # Build the project npm run build

Quick Start

As an MCP Server

Add to your MCP client configuration (e.g., Claude Desktop):

{ "mcpServers": { "consult": { "command": "node", "args": ["/path/to/mcp-consult/dist/index.js"], "env": { "OLLAMA_BASE_URL": "http://localhost:11434" } } } }

Running the Server

Make sure Ollama is running locally (default: http://localhost:11434).

# Start the server npm start # Or for development with auto-reload npm run dev # Run the demo client npm run demo

Usage Examples

Basic Consultation

// Request via MCP { "tool": "consult_ollama", "arguments": { "prompt": "Explain the benefits of microservices architecture", "model": "llama3.2", "context": "We're designing a new e-commerce platform" } }

Model Comparison

{ "tool": "compare_ollama_models", "arguments": { "prompt": "What are the trade-offs of NoSQL vs SQL databases?", "models": ["llama3.2", "qwen2.5-coder:7b", "deepseek-v3.1"] } }

Sequential Consultation Chain

{ "tool": "sequential_consultation_chain", "arguments": { "consultants": [ { "id": "analyst", "model": "llama3.2", "prompt": "Analyze the security implications of this API design: {code}", "timeoutMs": 120000 }, { "id": "architect", "model": "qwen2.5-coder:7b", "prompt": "Based on the analysis from {analyst}, suggest architectural improvements", "timeoutMs": 180000 } ] } }

Environment Variables

# Ollama server endpoint (default: http://localhost:11434) OLLAMA_BASE_URL=http://your-ollama-server:11434 # Memory storage directory (default: /tmp/mcp-consult-memory) MEMORY_DIR=/custom/memory/path # Memory MCP server configuration REMEMBER_MCP_CONFIG='{"type":"stdio","command":"node","args":["/path/to/memory-server.js"]}'

Memory Configuration

The remember_consult tool supports flexible memory backend configuration. It attempts to connect to memory storage in this order:

  1. REMEMBER_MCP_CONFIG environment variable (JSON config)

  2. VS Code mcp.json entries (auto-detects remember/memory servers)

  3. MEMORY_MCP_CMD/MEMORY_MCP_ARGS environment variables

  4. Local file fallback at MEMORY_DIR (default: /tmp/mcp-consult-memory)

Example Memory Server Configuration

# Use external memory MCP server REMEMBER_MCP_CONFIG='{"type":"stdio","command":"/usr/bin/node","args":["/path/to/memory-server.js"]}'

VS Code mcp.json (automatically detected):

{ "mcpServers": { "memory": { "command": "node", "args": ["/path/to/memory-server.js"] } } }

Tool Reference

consult_ollama

Consult with a specific Ollama model.

Parameters:

  • prompt (required): The consultation prompt

  • model (required): Ollama model name (e.g., "llama3.2")

  • context (optional): Additional context for the consultation

list_ollama_models

List all available models on the Ollama instance.

Parameters: None

compare_ollama_models

Run the same prompt against multiple models for comparison.

Parameters:

  • prompt (required): The prompt to send to all models

  • models (required): Array of model names to compare

  • context (optional): Shared context for all models

remember_consult

Store consultation results in persistent memory.

Parameters:

  • key (required): Unique identifier for the memory entry

  • value (required): Content to store

  • metadata (optional): Additional context about the stored data

sequential_consultation_chain

Execute multi-step reasoning chains where consultants build on previous responses.

Parameters:

  • consultants (required): Array of consultant configurations

    • id (required): Unique consultant identifier

    • model (required): Ollama model name

    • prompt (required): Consultation prompt (can reference previous consultants with {consultant_id})

    • timeoutMs (optional): Timeout in milliseconds (default: 120000)

Development

Project Structure

mcp-consult/ ā”œā”€ā”€ src/ ā”‚ ā”œā”€ā”€ index.ts # Main entry point ā”‚ ā”œā”€ā”€ handlers.ts # Tool request handlers ā”‚ ā”œā”€ā”€ services/ ā”‚ ā”‚ ā”œā”€ā”€ OllamaService.ts # Ollama API integration ā”‚ ā”‚ ā””ā”€ā”€ MemoryService.ts # Memory management ā”‚ ā”œā”€ā”€ utils/ ā”‚ ā””ā”€ā”€ types/ ā”œā”€ā”€ test/ # Test suites ā”œā”€ā”€ dist/ # Compiled output ā””ā”€ā”€ docs/ # Documentation

Running Tests

# Run all tests npm test # Run with coverage npm run test:coverage # Watch mode npm run test:watch # Run demo client npm run demo

Building

# Development build npm run build # Watch mode npm run build:watch

Sequential Consultation Chains

The sequential_consultation_chain tool enables complex multi-step reasoning by allowing consultants to reference and build upon previous responses. This creates powerful workflows for collaborative problem-solving.

Timeout Configuration

Configure timeouts based on task complexity:

{ "consultants": [ { "id": "analyzer", "model": "deepseek-v3.1:671b-cloud", "prompt": "Analyze this complex codebase...", "timeoutMs": 180000 } ] }

Recommended timeouts:

  • Simple queries: 60-90 seconds

  • Code generation: 180-300 seconds

  • Complex analysis: 300-600 seconds

Note: Avoid breaking complex questions into smaller parts, as this loses conversation context. Instead, increase the timeoutMs for consultants that need more processing time.

For detailed examples, see sequential_chain_demos.md.

Integration with Other MCP Tools

MCP Consult works seamlessly with other MCP servers:

  • mcp-optimist - Code optimization and analysis

  • mcp-tdd - Test-driven development workflows

  • Memory servers - Persistent data storage

  • Code analysis tools - Static analysis integration

Docker

Building the Image

FROM node:18-alpine WORKDIR /app COPY package*.json ./ RUN npm ci --only=production COPY dist/ ./dist/ CMD ["node", "dist/index.js"]
# Build and run docker build -t mcp-consult . docker run -p 3000:3000 mcp-consult

Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

Development Guidelines

  • Follow Test-Driven Development (TDD) practices

  • Maintain test coverage above 70%

  • Use TypeScript strict mode

  • Follow existing code style and formatting

  • Update documentation for new features

Architecture

For detailed technical architecture, see ARCHITECTURE.md.

License

MIT License - see LICENSE file for details.

Requirements

  • Node.js 18+

  • Ollama server running locally or accessible via HTTP

  • npm or pnpm for package management

Support

Links

Built with ā¤ļø using the Model Context Protocol and Test-Driven Development

One-click Deploy
A
security ā€“ no known vulnerabilities
A
license - permissive license
A
quality - confirmed to work

How are these scores calculated?

Resources

Tools

New MCP Servers

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/Atomic-Germ/mcp-consult'

If you have feedback or need assistance with the MCP directory API, please join our Discord server