MCP

The MCP server exposes read-only project memory tools to MCP-capable clients, including coding agents. Use MCP when a client supports structured tools. Use the CLI when a shell command is simpler or when you want direct human-readable output.

stdio server

Run a local stdio MCP server for clients that spawn tools as child processes (Codex, Claude Code, etc.):

memory mcp run    --project <slug>
memory mcp status --project <slug>

Stdio mode writes only MCP JSON-RPC messages to stdout; diagnostics go to stderr.

Streamable HTTP server

The service can expose MCP over a local HTTP route, typically /mcp. Use this only for local, token-bearing clients unless you add your own network and authentication controls.

Tools

The MCP surface is read-first and focused on project context.

ToolPurpose
memory_queryAsk a project question and return answers with citations.
memory_resumeGenerate a continuation briefing.
memory_up_to_speedGenerate a new-agent briefing.
memory_overviewInspect counts and project state.
memory_list_memoriesBrowse memory rows.
memory_get_memoryInspect one memory.
memory_memory_historyRead memory version history.
memory_list_activitiesInspect activity events.
memory_list_replacement_proposalsReview pending proposals without approving them.

Security

  • Prefer stdio for local desktop or CLI clients.
  • Keep Streamable HTTP bound to local interfaces unless you add network controls.
  • Require token-bearing clients for HTTP MCP.
  • Scope tools to explicit projects.
  • Do not expose MCP HTTP to the public internet.

Next

Read Agents, Watchers, or Operations.

© 2026 Olivier Van Acker (3vilM33pl3). Memory Layer is AGPL-3.0-or-later with commercial licensing available.

On this page