Mimir

Mimir is an MCP server that combines a persistent Neo4j graph database with AI embeddings to store tasks, contexts, and relationships for AI agents. It enables multi-agent orchestration and semantic search, giving your AI assistants a memory of past conversations and actions. Once running, you can interact with the MCP API to create, query, and relate tasks, contexts, and notes within a knowledge graph. The included web UI provides a portal for file indexing, an orchestration studio for visual workflows, and endpoints for chat and embeddings, allowing you to leverage memory-enabled reasoning in your AI workflows.

Prerequisites:

• Docker Desktop installed and running
• Node.js 18+ (optional if using the docker-based approach)
• Git installed

Installation steps (Docker-based):

1. Clone the repository:

git clone https://github.com/orneryd/Mimir.git
cd Mimir

2. Start services using Docker:

# Start all services in detached mode
docker compose up -d

3. Verify services:

# Check containers
docker ps
# Optional: view logs
docker logs <container_id>

4. Access the Mimir UI and MCP API:

• Web UI: http://localhost:9042
• MCP API: http://localhost:9042/mcp

If you prefer the npm-start workflow (local development):

1. Install dependencies and start:

# From repo root
npm install
npm run start

2. Validate status and logs:

npm run status
npm run logs

3. Access UI and APIs as above.

Tips and common considerations:

• Environment variables: HOST_WORKSPACE_ROOT is required for indexing and should point to your codebase directory. Tilde expansion (~) is supported.
• Neo4j password: set NEO4J_PASSWORD in production for security.
• If using the MCP REST API, ensure the base URL is reachable (default http://localhost:9042/mcp).
• For read-only workspace mounts in Docker, adjust your docker-compose volumes accordingly to add :ro.
• The system supports multiple LLM providers via MIMIR_DEFAULT_PROVIDER (e.g., openai, copilot, ollama). Configure embedding and model endpoints using the documented MIMIR_*_ENV variables.
• When indexing large codebases, index incrementally and monitor performance via the Orchestration Studio.
• If you encounter connectivity issues with the Embeddings or LLM endpoints, verify network access and endpoint paths in MIMIR_LLM_API and MIMIR_EMBEDDINGS_API.