Omega — Claude Code MCP server: memory across sessions
Omega — an MCP server that keeps decisions, lessons and task checkpoints for an agent. Semantic search and a relationship graph let a new session pick up where the last one stopped.
no license 3,525 installs updated 2026-02-15
claude mcp add --transport http omega-memory https://omega-memory--omegamemory.run.toolsWhat Omega does
Omega gives an AI coding agent persistent memory over HTTP. It records decisions, errors, lessons and task checkpoints, then retrieves them by semantic search or through a relationship graph. The point is simple: you stop explaining the same context in every new session.
The server exposes a set of tools under the omega_ prefix. omega_checkpoint captures the current plan, progress, touched files and decisions. omega_resume_task loads the latest checkpoint at the start of a new session. omega_traverse walks the graph from one memory to connected ones within N hops. omega_consolidate prunes stale entries and cleans orphaned edges, while omega_compact merges similar memories into summary nodes and marks the originals as superseded.
There are also reminders and stats. omega_remind sets a time-based reminder that fires at session start or during an active session. omega_weekly_digest returns a weekly summary with new memories, growth trends and top topics. omega_protocol is meant to be called right after omega_welcome at session start — it returns context-sensitive operating rules assembled from the project, peer activity and learned lessons.
For Claude Code the setup is a single command: claude mcp add --transport http omega https://omega-memory--omegamemory.run.tools. Codex uses codex mcp add omega --url https://omega-memory--omegamemory.run.tools. In Cursor and GitHub Copilot the server goes into the JSON config as an HTTP endpoint.
What it helps with
- resume a task in a new session without re-explaining the context
- discover related decisions through the memory graph instead of brute search
- save a checkpoint before the context window fills up
- set a time-based reminder that fires during the session
- clean stale entries and compress near-duplicate memories
Example prompts for the agent
- Call
omega_checkpointon this task and save the plan, progress and files I touched - Run
omega_welcomeat session start and tell me what we did last time - Use
omega_queryin semantic mode to find decisions about this error
How to install Omega
Platform CLI: remote server
claude mcp add --transport http omega-memory https://omega-memory--omegamemory.run.toolsPlatform CLI: remote server
codex mcp add omega --url https://omega-memory--omegamemory.run.toolsConfig file
{
"mcpServers": {
"omega": {
"url": "https://omega-memory--omegamemory.run.tools"
}
}
}Config file
{
"servers": {
"omega": {
"type": "http",
"url": "https://omega-memory--omegamemory.run.tools"
}
}
}Platform CLI: remote server
gemini mcp add -t http omega https://omega-memory--omegamemory.run.toolsStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Omega
How do I install Omega in Claude Code?
Run claude mcp add --transport http omega https://omega-memory--omegamemory.run.tools. The server connects over HTTP, no API key is needed.
What does Omega do?
It stores decisions, lessons and task checkpoints, then finds them through semantic search and a relationship graph. That is how an agent keeps context between sessions.
How is omega_checkpoint different from storing a memory?
A checkpoint captures a full task state: plan, progress, files and decisions. omega_resume_task restores that whole snapshot, not just individual entries.