Remote Graphify Server

A containerized implementation of Graphify exposing a Model Context Protocol (MCP) server over HTTP/SSE. This image is designed to be easily consumed by autonomous agents (like Hermes) via Docker Compose.

Features

• MCP Ready: Exposes Graphify's API over HTTP (SSE) for seamless agent integration.
• Multi-LLM Support: Pre-configured environment variables for Ollama, Anthropic, Gemini, and OpenAI.
• IPv6 Enabled: Configured for modern networking architectures with IPv6 support.
• Pre-installed Dependencies: Built with uv and includes graphifyy[all] for maximum compatibility out-of-the-box.

Quick Start

1. Extract your Graph

The MCP server reads from a static graph file. Before starting the server, you must generate the graph.json file from your project code. Run this locally or via an ephemeral container inside your project directory:

uvx graphifyy extract

This will create a graphify-out/ directory containing the required data.

2. Start the Server

Use the provided docker-compose.yml to bring up the service:

docker compose up -d

3. Connect your Agent

Configure your agent (e.g., Hermes) to connect to the remote MCP server using the SSE transport. If both containers share the same Docker network, use the internal hostname:

• Endpoint URL: http://graphify:8080/sse

Configuration Options

Modify the environment section in your docker-compose.yml to configure your preferred LLM provider.

Local Models (Ollama)

If you are running Ollama on your Docker host machine, be sure to use host.docker.internal or your host's local IP address instead of localhost:

OLLAMA_BASE_URL=[http://host.docker.internal:11434](http://host.docker.internal:11434)
GRAPHIFY_OLLAMA_KEEP_ALIVE=10

Cloud Models

Uncomment and populate the relevant variables for cloud providers:

• ANTHROPIC_API_KEY / ANTHROPIC_MODEL
• GEMINI_API_KEY
• OPENAI_API_KEY / OPENAI_BASE_URL / OPENAI_MODEL

---

Docker Hub: sinfallas/remote-graphify