Claude Code plugin

The agentmemory plugin packages the remote MCP connection for Claude Code. Once installed, Claude Code has the Agent Memory tools (search_memory, add_memory, consolidate, report_outcome and the rest) and receives the server's memory-usage instructions on every session.

The plugin is a thin wrapper: it registers one HTTP MCP server named agent-memory and reads its URL and API key from your environment. It holds no code of its own.

InstallCopy Anchor Link

export AGENT_MEMORY_API_KEY="amk_..."
# Only for a self-hosted deployment:
# export AGENT_MEMORY_URL="https://memory.example.internal/v1/mcp"

/plugin marketplace add erkansezgin/agentmemory-claude-plugin

/plugin marketplace add /path/to/agentmemory-claude-plugin

/plugin install agentmemory@agentmemory

What it configuresCopy Anchor Link

The plugin's MCP configuration is equivalent to:

{
  "agent-memory": {
    "type": "http",
    "url": "${AGENT_MEMORY_URL:-https://memory.empecta.com/v1/mcp}",
    "headers": {
      "Authorization": "Bearer ${AGENT_MEMORY_API_KEY}"
    }
  }
}

If you prefer not to use a plugin, the same connection is one command:

claude mcp add --transport http agent-memory "$AGENT_MEMORY_URL" \
  --header "Authorization: Bearer $AGENT_MEMORY_API_KEY"

TipsCopy Anchor Link

• Pick the key's agent binding deliberately. Create the key with a dedicated agent ID such as claude-code, so memories written from Claude Code are attributed to it. The tools never need an agentId argument unless the key is bound to several agents.
• Scope per project. Point different projects at different workspaces (separate keys) or at application-scoped keys, so each codebase recalls its own decisions plus the shared workspace pool. See Scoping.
• Reinforce the loop. Claude Code receives the server's usage protocol automatically. For extra reliability, add the memory rule from the MCP guide to your project's CLAUDE.md.
• Keys stay out of the repo. The plugin only references environment variables; never commit an amk_ key.