Memory

You are setting up a Memory Server MCP that stores text passages and retrieves them by semantic similarity. This lets you memorize multiple texts and later query for the most relevant stored content, powered by vector storage and embedding models.

How to use

You interact with this Memory Server through an MCP client to memorize texts and later retrieve the most relevant passages for your questions.

Memorize a single text by asking the MCP to store it. For example, you can say or send: Memorize this text: "Singapore is an island country in Southeast Asia." After the MCP processes it, the text is stored for semantic retrieval.

Memorize multiple texts at once by providing a list of passages. For example: Memorize these texts: - Singapore is an island country in Southeast Asia. - It is about one degree of latitude north of the equator. - It is a major financial and shipping hub. The server stores all provided texts for later semantic retrieval.

You can memorize the contents of a PDF file by using the dedicated memorize_pdf_file tool. The MCP reads up to 20 pages at a time, extracts text, chunks it into meaningful segments, and stores those chunks. You can start from a specific page if needed, and the process repeats until the entire document is memorized.

To retrieve information, ask a question like: What is Singapore? The MCP returns the most relevant stored texts along with descriptions explaining their relevance.

How to install

# Prerequisites
# Install uv, a fast Python package manager
curl -LsSf https://astral.sh/uv/install.sh | sh

# If you are on Windows, you can install uv via PowerShell
# powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

# Obtain the Memory Server project and navigate into it
# (Clone not shown here; follow your usual workflow for obtaining the Memory Server project)

# Start the auxiliary services with Docker Compose
docker-compose up

# After containers start, pull the embedding model for Ollama
docker exec -it ollama ollama pull all-minilm:l6-v2

# Configure MCP server to point to this local Memory Server
# The exact path will depend on your setup; see the provided JSON snippet for the server config

"mcp-rag-local": {
  "command": "uv",
  "args": [
    "--directory",
    "path\\to\\mcp-rag-local",
    "run",
    "main.py"
  ],
  "env": {
    "CHROMADB_PORT": "8321",
    "OLLAMA_PORT": "11434"
  }
}

# Start the MCP server locally using the configured command block above
# The final start command is the one you would run from your environment

Additional setup notes

A web-based GUI is available to inspect and manage the vector database. Open http://localhost:8322 in your browser to view and manage memory stored in ChromaDB.

Environment variables used by the local MCP server are CHROMADB_PORT and OLLAMA_PORT. You can adjust these ports if needed, but ensure your Docker Compose, Ollama, and MCP server configuration align with the chosen values.

Tools and capabilities

The Memory Server exposes the following MCP tools for interacting with your memory store:

• memorize_multiple_texts: Store several texts at once for semantic retrieval.

• memorize_pdf_file: Read a PDF, extract text in chunks, and memorize the segments.

Security and access considerations

When running locally, limit access to the Memory Server’s endpoints to trusted clients. Use network-level controls and, if needed, add authentication in front of the MCP layer to prevent unauthorized memorization or retrieval.

Troubleshooting

If the memory store fails to populate or retrieval returns nothing, verify that the Docker containers are running, the Ollama model is pulled, and the MCP server process is active with the correct environment variables for ports.

Available tools

memorize_multiple_texts

Stores multiple texts in the memory store for later semantic retrieval.

memorize_pdf_file

Reads a PDF, extracts text in chunks, and memorizes those chunks for semantic retrieval.