Oceanir Memory

Oceanir Memory provides a persistent memory store for AI coding agents, enabling you to remember context across sessions, store entities like people and projects, link relations, add observations, and search and explore connected knowledge.

How to use

To connect with Oceanir Memory from your MCP client, you will typically configure two endpoints: a local API server for memory storage and a local MCP interface to run the memory server alongside your agent workflow.

You can store memories, search through them, observe changes to entities, and relate different memories to reflect your coding context. Your MCP client will send remember, recall, observe, relate, and memory-related commands to the configured endpoints, enabling seamless cross-session continuity for your AI agents.

How to install

Prerequisites: you need Node.js and npm installed on your machine.

Install dependencies and build the server package using npm, then start the API server and connect your MCP client.

Recommended commands to set up locally:

# Install dependencies and build
npm install
npm run build

# Start the API server (provides memory storage endpoints)
npm run api
# Runs on http://localhost:3000

# Add this MCP config to your client (examples shown below)
# The exact path may vary depending on your MCP client setup

Additional configuration and usage notes

MCP clients can connect to Oceanir Memory via two explicit configurations shown in the usage examples: a remote HTTP API endpoint and a local MCP server you run with a simple node command.

Tools and endpoints

Oceanir Memory exposes several functions that your MCP client can invoke to manage memories and observations. The available tools include: remember to store a memory, recall to search memories, observe to add an observation to an entity, relate to create a relation between entities, memory_stats to get server statistics, get_preferences to retrieve all preferences, get_patterns to fetch patterns and solutions, and forget to delete an entity.

Security and best practices

Keep your memory store accessible only to trusted MCP clients. If you expose the API publicly, implement authentication and restricted access controls to protect sensitive data such as personal information, project details, and observed notes.

Troubleshooting

If you cannot reach the API at http://localhost:3000, verify that the API server is running and listening on the expected port. Check server logs for startup messages and confirm network access from your MCP client to that host and port.

Notes on MCP integration

In MCP configurations, you can point to Oceanir Memory in two ways: using an HTTP endpoint for remote interaction or using a local stdio connection when running the MCP server alongside your agent.

Available tools

remember

Store a memory with a name, type, and content to persist across sessions.

recall

Search memories using a query string to retrieve matching results.

observe

Add an observation note to a specific entity to track changes over time.

relate

Create a relationship between two entities to reflect connections like uses, prefers, or solved.

get_preferences

Retrieve all stored user or system preferences.

get_patterns

Fetch known patterns and their associated solutions.

memory_stats

Get statistics about the stored memories and usage.

forget

Delete an entity from memory.