Plugins

Plugins

The fastest way to give Claude Code and Codex a memory that works every turn

The HyperMemory plugins install everything an agent needs in one step: the MCP connection, the HyperMemory skill, the hooks that make the agent use memory on every turn, and a background memory-writer agent that saves what it learned.

Connecting the MCP server alone gives an agent the tools. The plugin makes it use them — reliably, without you having to say “remember this”.

Each plugin installs with one instruction you paste into the agent — it handles the terminal commands, the sign-in and the skill for you.

Using another tool, or want a plugin tailored to your team? See Customized Plugins.

What a plugin does on every turn

1

Recall before replying

At the start of a session the agent calls hm_get_overview, and before every reply it calls hm_recall with the words of your message — including your organization's memory if you belong to one.

2

Answer with your context

The agent uses what it recalled: past decisions, preferences, project history, the people involved.

3

Save what it learned

At the end of the turn the agent hands a short summary to the memory-writer, a background agent that stores new facts, updates outdated ones, links them, writes a timeline entry and reports the turn's token usage. You never wait for it.

What’s included

ComponentClaude CodeCodex
MCP connection to https://api.hypermemory.io/mcp (OAuth)✓✓
hypermemory skill — the every-turn protocol✓✓
Hooks that load the skill and enforce recall and the writer✓✓
memory-writer agent✓✓
memorize-full-chat skill — save a whole conversation on demand✓—
Exact token counts from the clientEstimatedExact
Organization memory and per-project organization binding✓✓

Other agents

Not using Claude Code or Codex?