MCP Ollama Consult Server
# MCP Ollama Consult Server
[](https://www.typescriptlang.org/)
[](https://modelcontextprotocol.io)
[](https://github.com/Atomic-Germ/mcp-consult/actions/workflows/ci.yml)
[](https://glama.ai/mcp/servers/@Atomic-Germ/mcp-consult)
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
- **compare_ollama_models**: Run the ame prompt against multiple Ollama models and return their outputs side-by-side for comparison
- **remember_consult**: Store the result of a consult into a local memory store (or configured memory service)
## Installation
1. Install the server:
```bash
npm i -g https://github.com/Atomic-Germ/mcp-consult/releases/download/v1.0.1/mcp-ollama-consult-1.0.1.tgz
```
2. Configure the server:
```json
{
"servers": {
"ollama-consult": {
"type": "stdio",
"command": "mcp-ollama-consult",
"args": []
}
},
"inputs": []
}
```
## Usage
Make sure Ollama is running locally (default: ([http://localhost:11434](http://localhost:11434)).
Start the MCP server:
```bash
mcp-ollama-consult
```
Or for development:
```bash
npm run dev
```
## Configuration
Set the `OLLAMA_BASE_URL` environment variable to change the Ollama endpoint:
```bash
OLLAMA_BASE_URL=http://your-ollama-server:11434 npm start
```
## Docker
To run with Docker, build the image:
```dockerfile
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
TDQS
Scored across 5 tools
The purposes of list_ollama_models and remember_context are distinct, but consult_ollama, compare_ollama_responses, and sequential_consultation_chain heavily overlap. All three involve querying models, differing mainly in whether responses are combined or chained, which could lead to agent confusion.
Tools mostly follow a snake_case verb_noun pattern (list_ollama_models, remember_context, compare_ollama_responses), but 'sequential_consultation_chain' breaks the verb-first pattern with an adjective-noun structure. Also, 'consult_ollama' is a general verb that overlaps with others, making the naming less distinctive.
With only 5 tools, the server is well-scoped and each tool seems to have its place, though the overlap among core consultation tools slightly reduces the necessity of having three separate ones. Still, it's a small, manageable set that fits the server's purpose.
The server covers a typical consultation workflow: listing available models, querying them, comparing responses, chaining consultations, and maintaining context. A minor gap is the lack of a tool to explicitly clear context or manage sessions, which agents might need for long-running interactions.