Connect Membrick to your AI agent¶

Membrick can give supported AI agents memory that carries across work sessions. Install it once for a project and the agent can look up earlier decisions and save useful outcomes as it works. Memories from connected agents are kept in the project's .membrick/memory folder.

Choose your agent¶

Install for Codex, Claude Code, Gemini CLI, or Antigravity¶

Open a terminal in the project folder and run the command for your agent. The installer downloads and checksum-verifies the official Membrick release, configures the selected agent, and checks that its MCP connection works.

# Codex
curl -fsSL https://raw.githubusercontent.com/RooAGI/Membrick/main/scripts/install.sh | sh -s -- --agent codex --project .

# Claude Code
curl -fsSL https://raw.githubusercontent.com/RooAGI/Membrick/main/scripts/install.sh | sh -s -- --agent claude --project .

# Gemini CLI
curl -fsSL https://raw.githubusercontent.com/RooAGI/Membrick/main/scripts/install.sh | sh -s -- --agent gemini --project .

# Antigravity CLI
curl -fsSL https://raw.githubusercontent.com/RooAGI/Membrick/main/scripts/install.sh | sh -s -- --agent agy --project .

Restart the agent if it was already open. Then ask a question about a past project decision, or ask the agent to remember an outcome from the current session. The agent uses Membrick's memory tools and lifecycle integration in the background.

If you already installed the membrick executable, use the shared provider commands for setup and maintenance:

membrick install codex .
membrick install claude .
membrick install gemini .
membrick install agy .
membrick serve codex .
membrick verify codex .
membrick uninstall codex .

Use membrick hook <provider> <event> [PATH] to run a lifecycle hook, such as membrick hook claude session-start .. setup remains an alias for install. Older provider-specific flags remain accepted for installed agent configs and existing scripts.

Advanced path overrides use --provider-config, --provider-settings, --provider-skills-dir, --provider-hooks-dir, and --provider-plugin-dir. Use --force-skill with install when you intentionally want to replace a customized skill.

Uninstall Claude Code or Codex¶

From the project directory, remove the Membrick integration with:

membrick uninstall claude .
membrick uninstall codex .

Uninstall removes only Membrick or legacy Lint-AI MCP entries, hooks, and project memory guidance for that provider. It preserves unrelated agent settings, customized Claude skills, the project's .membrick memory data, and the Membrick executable. Restart the agent after uninstalling. The Codex hook feature flag is left unchanged because other Codex hooks may use it.

Migrate from Lint-AI¶

Existing Lint-AI users can install Membrick and update one agent connection in one step. From the project directory, run:

curl -fsSL https://raw.githubusercontent.com/RooAGI/Membrick/main/scripts/migrate-lint-ai.sh | sh -s -- --agent codex --project .

Replace codex with claude, gemini, or agy for another supported agent. The migration installs Membrick, updates that agent's configuration while preserving unrelated settings, and verifies the MCP connection. During agent setup, existing .lint-ai project state is moved to .membrick with its contents preserved. Membrick also checks for legacy state when it first accesses the project. If both directories exist, migration merges non-conflicting entries and moves conflicting legacy entries into .membrick/migration-archive, without overwriting current state. It removes the project-root .lint-ai directory when migration completes. The old lint-ai executable is left installed; once you have confirmed Membrick works, remove it using the package manager that installed it.

OpenClaw and Hermes¶

OpenClaw and Hermes have their own setup steps because each host loads its integrations differently. Their guides walk through the install and explain what is automatic:

• Set up OpenClaw: install the MCP connection, memory skill, recall hook, and capture plugin, then restart the Gateway.
• Set up Hermes: install the MCP connection and memory skill. The optional hooks plugin adds automatic recall and capture and requires the Membrick server to be running.

Both integrations use the same project memory folder, .membrick/memory, as the other agents. The agents do not need to be switched to a different memory provider.

What happens after setup¶

• The agent can search project memory when it needs earlier context.
• Supported lifecycle integrations can recall relevant memories at the start of work and capture outcomes when work finishes.
• You can add or remove memories yourself by editing files in .membrick/memory.
• You can turn memory behavior on or off with the agent's Membrick controls; this does not delete saved memories.

See Quickstart to try memory directly from the command line, or Agent integrations for the implementation details and shared memory behavior.