Skip to content
harbor
How it worksAgentsReviewMeasureDocsPricing
Get started →
How it worksAgentsReviewMeasureDocsPricingGet started →
Home/Blog

Blog

  • Claude Code and AGENTS.md
  • Claude Code skills
  • Claude Code hooks
  • claude-mem
  • Claude Code best practices
  • Karpathy CLAUDE.md
  • AGENTS.md vs CLAUDE.md
  • Cursor rules
  • copilot-instructions.md
  • Why Claude ignores it
  • CLAUDE.md length
  • Committing CLAUDE.md
  • Agent memory compared
  • CLAUDE.md vs skills vs hooks
  • Same mistake again
  • Review comments to rules
  • Slack decisions
  • Company brain
  • Context engineering, Claude 5
  • Context rot, stale rules
  • AGENTS.md research
  • ADRs for agents
  • Memory poisoning
  • MCP memory servers
  • Claude Code, Cursor, Codex
  • Growing CLAUDE.md
  • Served vs cited

Blog

Share memory between Claude Code, Cursor and Codex

Published: September 10, 2026

You have Claude Code in one terminal. Cursor in the editor. Codex on the pull request. You told Claude the refund window is 60 days on annual plans. Cursor did not hear you. Codex will invent 30.

That is the whole product category, described from the chair. Each agent has a memory. None of them has the others'. The thread at anthropics/claude-code#38536 is people asking for shared team memory for Claude Code. The same hole exists between Claude Code and everything else on the desk.

The usual fix is a file. CLAUDE.md, AGENTS.md, a pasted brief. It works until it is in three places, then four, and the one you edited last week is not the one Cursor read this morning.

Harbor is one graph. The CLI puts it on the machine. The MCP server puts it in the chat. Slack and the web app are the same graph for people who will not open a terminal. This note is about the two that sit next to Claude Code, Cursor and Codex.

The CLI

On the machine:

npm install -g harborloop
harbor init

harbor init installs the hooks those three agents already know how to run. After that you do not prefix a command. You do not wrap the agent. You run Claude Code the way you already run it. Harbor is already on the machine, and the session starts with the facts that apply to this repo, this task.

harbor doctor is the check that recall is reaching the agents you think it is reaching.

The MCP server

If the agent speaks MCP, it does not need the CLI. It calls Harbor the way it calls anything else: search, then fetch the statement. The answer comes back with a [#1] on the line it used. An answer without a citation is labelled a guess inside the product; this page has no business being vaguer than that.

Cursor, Claude Code and Codex can all sit on the same graph this way. What one session learns goes through review before any of them see it. Nothing reaches an agent that you did not allow.

What this is not

This is not a shared chat transcript. It is not a dump of every Slack message into the prompt. Harbor picks the two lines that apply and leaves the rest out. The next note is why that matters once the file you were going to share starts to grow.

harbor

Decide once. Every agent knows. One company brain for all the agents your team runs, built from work you already do and kept only while it is still true.

Product

  • How it works
  • Review
  • Your agents
  • Guardrails
  • What it counts
  • Pricing

Developers

  • Docs
  • Blog
  • CLI
  • MCP server
  • Served and cited
  • Environments

Company

  • Get started
  • Contact
  • Privacy Policy
  • Terms of Service
  • Data Processing Agreement
  • Refund Policy
© 2026 harbor·Product names and logos are trademarks of their respective owners.