HOW IT WORKSCONNECT YOUR TOOLS
Back to Connect
Connect

Install the Hark mod for Claude Code

Install the Hark mod so Claude Code reads your project brief at session start and proposes what it learned for you to approve at the end.

MCP endpoint
https://harkstudio.io/mcp

Sign in to Hark to connect a client. Personal tokens and per-account details live inside the app.

The Hark mod for Claude Code (Early access)

A Claude Code plugin (hark-memory): starts every session briefed, keeps the plan through compaction, guards agreed decisions, and ends with a handoff. Install with /plugin install hark-memory@hark; pair using a one-time code from Connections.

Early access. The Claude plugin directory listing is pending; the install from the repo below works today.

What it does

1. Brief on start. When a session starts, the mod fetches the project's compact brief and adds it to your first message, so Claude already knows the plan, the agreed decisions and the next action.

2. Survives compaction. When Claude Code compacts the conversation, the agreed decisions, non-goals, blockers and next action come back right after, so the plan isn't lost halfway through a long session.

3. Decision guard. Before Claude edits a file covered by an agreed decision or non-goal, you see the decision and why, and choose Allow or Stop. The check runs on your machine; nothing about the edit is sent.

4. Stop = handoff. When the session ends, Hark gets a short summary — files touched, commits, test results and Claude's closing notes — and drafts the handoff for you to review. Sessions with no changes write no handoff.

5. Status line and /hark. The status line shows the project, its handoff state and how many items need you. /hark brief, /hark handoff, /hark needs and /hark open do the obvious things.

Hark tools the mod's Claude can call: get_agent_brief, start_session, end_session, add_journal_entry, update_build_state, list_candidates, create_handoff (makes a handoff link for a developer, client or agent) and the other Hark tools on the same connection.

Install

From the Claude plugin directory (works once the listing is live):

/plugin install hark-memory@claude-plugins-official

Works now — from the repo, in a Claude Code session:

/plugin marketplace add Hark-Studio/hark-mod
/plugin install hark-memory@hark

Or from a shell:

claude plugin marketplace add Hark-Studio/hark-mod
claude plugin install hark-memory@hark

Setup

  1. Pairing: Connections → Access keys → Claude Code preset → Create key. Copy the setup message containing an 8-character code, valid once for 15 minutes. The agent exchanges it at POST https://harkstudio.io/api/v1/pair with { code } (no auth), then stores access_key privately in its sensitive configuration. Never paste the key into chat. Existing plugin versions may need the exchange performed by the agent first.
  2. Project: the repo says which Hark project it is with a file named .hark/project holding the project's code (or a .hark file holding venture=V-012). One line does it:
mkdir -p .hark && echo V-012 > .hark/project

Confirm it works: start Claude Code in the project and run /hark brief. Without a project file or a key, the mod stays idle and sends nothing.

Upgrading from 0.2.0

The plugin was renamed from hark to hark-memory in 0.2.1. Uninstall the old one, refresh the marketplace, install the new one, then pair again using a fresh code:

/plugin uninstall hark@hark
/plugin marketplace update hark
/plugin install hark-memory@hark
/plugin configure hark-memory@hark

Requirements

  • Claude Code 2.1.287 or newer.
  • A pairing code — open Connections → Access keys and select the Claude Code preset.
  • The project linked in the repo: a .hark/project file with the project's code (or a .hark file). Without one, the mod stays idle and sends nothing.
Create key for Claude Code →

Configuration

  • Access key: the plugin's access_key setting, marked sensitive. Claude Code keeps it in its own secure credential store (not settings.json) and hands it to the plugin; the mod never reads environment variables or anything else for the key. Change it with /plugin configure hark-memory@hark.
  • Project: the mod reads .hark/project, or a .hark file holding venture=V-012. It looks in the folder Claude Code starts in, then each parent folder, so monorepo packages work.
  • Turn off single behaviors (the decision guard, the re-inject after compaction, the status line) in the plugin's settings in Claude Code's /plugin menu.

Privacy and trust

  • The only file the mod writes is ~/.claude/hark/<id>.json (its own state, in a folder only you can read).
  • The only network address it contacts is https://harkstudio.io/mcp. /hark open also opens your project's page on harkstudio.io in your browser.
  • Sent: the project code, brief requests, and at the end a handoff summary — file paths, commit subjects, test pass/fail, and Claude's closing notes.
  • Never sent: file contents, diffs, your prompts, the transcript, tool output, or environment variables. Your access key only travels in the Authorization header.
  • Closing notes are cleaned before sending — code blocks are dropped and known secret formats are replaced with [redacted] — but that cleaning is best effort. A secret written out in plain prose could be sent. The README says the same.
  • Every request is abandoned after 3 seconds; nothing ever blocks a prompt, an edit or an exit. Claude Code plugins run unsandboxed, so this one is a single small, dependency-free file you can read before installing.
Read the source: github.com/Hark-Studio/hark-mod

Other tools

The mod is Claude Code only. Cursor and Codex keep using their install buttons and Hark's hooks, and work alongside it.

Troubleshooting

  • Brief not added: check the key is set (/plugin configure hark-memory@hark), the repo has .hark/project or .hark, and claude --version is 2.1.287 or newer.
  • Guard asks too often: give decisions a scope (an area like billing, or a path like src/db/**) so they only cover the files they're about.
  • No handoff drafted: usually the network at exit. Hark drafts one from repo activity after 45 minutes idle, so nothing is lost.
  • Uninstall: /plugin uninstall hark-memory@hark (and /plugin marketplace remove hark).

Verify it works

Ask your client: "List my Hark workspaces." Expected: your workspace names.

Then: "Create a project called Connection test in <workspace>." Expected: a new V-### appears in the app and the activity log shows the source ("via Claude Code", "via Codex", etc.). Delete the test venture afterward.

Troubleshooting

  • "Unauthorized request origin" — the client origin isn't allowlisted; retry from the latest app version, or email hello@harkstudio.io.
  • Consent screen loops / stale token — remove the connector in the client, sign out of Hark, sign back in, re-add the connector.
  • Tools list is empty — the connector isn't enabled for this chat (Claude and ChatGPT require enabling it per conversation).
  • Plan cap error — you've hit the monthly agent-call cap for this workspace; see billing.
  • Old endpoint — anything pointing at a lovable.app URL must be updated to harkstudio.io/mcp.

What Hark keeps clear

Evidence on every line

Every record shows Agreed / Observed / Proposed and a source you can open: where did this come from? Agreed means accepted, not independently verified. Observed is reserved for system evidence such as a merged pull request. An agent's claim is never shown as verified simply because it was accepted.

Learn more about Evidence on every line

Repo-native rules

Once linked and selected as the source, the repo's CLAUDE.md or AGENTS.md is authoritative. Hark mirrors it on pushes and retains history rather than authoring competing edits. Optional .hark/ export carries accepted handoff state and decisions into the repo. Repo activity suggests drafts with commit evidence. Agent briefs include mirrored instruction documents and linked repo metadata. Arbitrary repo-document browsing is not exposed by the current tools: Hark never serves source code. Automatic session hooks and compaction protection in the Hark mod are Claude Code-only; other tools follow their connection guide and explicit session tools.

Learn more about Repo-native rules

Scoped, expiring agent access

New agent keys default Propose: suggest changes, never approve them. An eight-character pairing code expires after 15 minutes and works once; no key is pasted into a setup chat. One-link agent approval asks you to sign in, choose permissions, projects and expiry, then approve. The agent still completes pairing. Keys normally expire after 90 days and can be revoked or rotated. The separate show-once Copy key path is for private storage, not chat.

Learn more about Scoped, expiring agent access
SIGN IN TO CONNECT