Share memory across Claude Code, Codex, and Cursor

Claude Code has its own conversation history. Cursor has its own. Codex, ChatGPT, VS Code, Windsurf, Gemini CLI, and any headless script you run each keep their own idea of what's going on, if they keep one at all. Context doesn't move between them on its own — it moves through you, copying and pasting a summary from one chat into the next.

That's fine for a single afternoon. It falls apart across a real project: you switch tools because one is better at a task, you come back after a week away, or a teammate picks up where you left off in a different client entirely. Nobody has the full picture, including you.

Hark runs a hosted MCP hub at harkstudio.io/mcp. Connect any MCP-capable client to it once and it can read and write the same project — the project page, the Current state, the journal, the decisions log — that every other client sees. See the full

setup steps for each client →

• Business, Build docs, and Branding tabs hold the thesis, stack, scope, and AI context doc an agent needs before it writes a line of code.
• The Current state card is the single place phase, summary, blockers, and next steps live, so an agent doesn't have to reconstruct status from git log.
• The journal records ships, decisions, sessions, and milestones with the source of each entry, including which agent made it.

Hark's session flow is three calls: start_session and get_agent_brief when a session opens, writes during the session, and end_session when it wraps, with what it did, what changed, what's next, and what to watch out for. That closing note is what the next agent — or you, tomorrow — reads before doing anything else.

The one-command installer writes the plumbing for this into your project: a Claude Code SessionStart hook, a stop reminder, a Cursor rule, an AGENTS.md block, and a .hark file. Most agents pick up the session flow without you having to explain it every time.

The reason you re-explain the same trade-off to three different tools is that the reasoning behind it never left the conversation it happened in. Hark's journal keeps decisions as first-class entries next to ships and milestones, so 'why we chose X over Y' is something any agent can look up instead of something only you remember.

None of this requires giving up your favourite tool. Hark doesn't replace Claude Code, Cursor, or Codex — it's the shared notebook they all write to. See how the rest of the pipeline fits together on the

how it works page →