MCP Ollama Consult Server

An MCP (Model Context Protocol) server that allows consulting with Ollama models for reasoning from alternative viewpoints.

Features

• consult_ollama: Send prompts to Ollama models and get responses
• list_ollama_models: List available models on the local Ollama instance

Installation

1. Ensure you have Node.js installed
2. Install dependencies and build:

npm i
npm run build

Usage

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

Start the MCP server:

npm start

Or for development:

npm run dev

Configuration

Set the OLLAMA_BASE_URL environment variable to change the Ollama endpoint:

OLLAMA_BASE_URL=http://your-ollama-server:11434 npm start

Memory (remember_consult) configuration

The server implements a remember_consult tool that will try, in order:

1. REMEMBER_MCP_CONFIG environment variable (JSON config or simple command string)
2. VS Code mcp.json entries (looks for a server key containing remember then memory)
3. MEMORY_MCP_CMD / MEMORY_MCP_ARGS environment variables
4. Falls back to a local file store at MEMORY_DIR or /tmp/mcp-consult-memory

Examples:

Use a simple stdio command via env (no shell quoting here):

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

Or point to a memory server described in your VS Code mcp.json (the server keys remember/memory are automatically detected).

If no memory MCP is available, the tool will write JSON observations to MEMORY_DIR (default /tmp/mcp-consult-memory).

Demo client

There is a simple demo client that spawns the server over stdio and exercises the tools (list, consult, compare, remember):

npm run build
npm run demo

Tests & CI

Unit tests (Vitest) and a GitHub Actions workflow were added. Run tests locally with:

npm install
npm test

Docker

To run with Docker, build the image:

FROM node:18-alpine
WORKDIR /app
COPY package*.json ./
RUN npm ci --only=production
COPY dist/ ./dist/
CMD ["node", "dist/index.js"]

Requirements

• Node.js 18+
• Ollama running locally or accessible via HTTP