@arcaelas/rag

MCP server with Ollama + Vectra for semantic memory and RAG operations.

Build intelligent AI agents with persistent semantic memory - Store, search, and retrieve knowledge using vector embeddings, powered by local Ollama models and Vectra vector database.

Features

• 🧠 Semantic memory with vector embeddings
• 🔍 Semantic search using similarity
• 📦 Bulk import/export via JSONL
• 🚀 Local-first with Ollama and Vectra
• 🔧 Zero configuration with sensible defaults

Prerequisites

• Node.js >= 18
• Ollama running locally
• An embedding model installed (e.g., ollama pull nomic-embed-text)

Installation

Using npx (recommended)

Add to your ~/.claude.json:

{
  "mcpServers": {
    "rag": {
      "command": "npx",
      "args": ["-y", "@arcaelas/rag"],
      "env": {
        "OLLAMA_HOSTNAME": "http://localhost:11434",
        "OLLAMA_MODEL_NAME": "nomic-embed-text"
      }
    }
  }
}

Global installation

npm install -g @arcaelas/rag

# Or with yarn
yarn global add @arcaelas/rag

Then in ~/.claude.json:

{
  "mcpServers": {
    "rag": {
      "command": "rag",
      "args": [],
      "env": {
        "OLLAMA_HOSTNAME": "http://localhost:11434",
        "OLLAMA_MODEL_NAME": "nomic-embed-text"
      }
    }
  }
}

Environment Variables

| Variable | Default | Description | |----------|---------|-------------| | OLLAMA_HOSTNAME | http://localhost:11434 | Ollama server URL | | OLLAMA_MODEL_NAME | nomic-embed-text | Embedding model name |

Available Tools

save(content, metadata?)

Save knowledge to semantic memory database.

await save("TypeScript is a typed superset of JavaScript", {
  category: "programming",
  importance: "high"
})

search(query, limit?)

Search knowledge base using semantic similarity.

await search("typed javascript", 5)

list(offset?, limit?)

List stored memories with pagination.

await list(0, 10)

get(id)

Retrieve specific memory by ID.

await get("uuid-here")

tag(id, tag)

Add tag to a memory.

await tag("uuid-here", "important")

destroy(id)

Permanently delete a memory.

await destroy("uuid-here")

upload(filename)

Bulk import memories from JSONL file.

await upload("/path/to/memories.jsonl")
// Returns: { filename, done: 299, error: [{ line: 297, error: "..." }] }

download(offset?, limit?, filename?)

Export memories to JSONL file.

await download(0, 100, "/path/to/export.jsonl")
// Returns: { filename, offset, limit, count: 303 }

JSONL Format

Each line in the JSONL file should be:

{"context": "Your knowledge content here"}

Optional fields:

{
  "context": "Content here",
  "metadata": {
    "category": "programming",
    "importance": "high",
    "project": "my-project"
  },
  "tags": "tag1,tag2,tag3"
}

Data Storage

Vector database is stored in:

• npx/global install: ~/.cache/@arcaelas/rag/data/
• Local install: <project-root>/data/

Collection name: arcaelas_mcp_rag_collection

Development

# Clone repository
git clone https://github.com/arcaelas/rag.git
cd rag

# Install dependencies
yarn install

# Build
yarn build

# Run locally
yarn start

# Watch mode
yarn dev

Contributing

Contributions are welcome! Please read our contributing guidelines before submitting PRs.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'feat: add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Security

See SECURITY.md for security policies and reporting vulnerabilities.

Changelog

See CHANGELOG.md for release history.

License

MIT © Miguel Guevara (Arcaela)

Links

• npm Package
• GitHub Repository
• Issues & Bug Reports
• Model Context Protocol
• Ollama

Support

• 📧 Email: [email protected]
• 🐛 Issues: GitHub Issues
• 💬 Discussions: GitHub Discussions