@memloom/mcp

The MCP server for memloom, a local-first memory engine for AI agents. Gives your agent a memory it can write to and search, shared with every other tool on your machine.

Setup

Add to your MCP client's config:

{
  "mcpServers": {
    "memloom": { "command": "npx", "args": ["-y", "@memloom/mcp"] }
  }
}

For Claude Code:

claude mcp add memloom -- npx -y @memloom/mcp

Nothing else to install. The server routes through the memloom serve daemon and starts it if needed, so agents, the CLI and the viewer share one store with no lock conflicts.

The tools

| Group | Tools | | --- | --- | | Memories | save_memory, recall_memory, read_passage, memory_history | | Source material | add_link, add_file, list_documents | | Decisions | list_conflicts, resolve_conflict, list_possible_contradictions, answer_possible_contradiction | | Graph | related_entities, set_schema_entry_status, delete_schema_entry |

recall_memory searches saved memories and ingested documents together, by meaning, exact words and entities, and returns sources with each result. save_memory dedupes and flags contradictions rather than overwriting.

Configuration

Put OPENROUTER_API_KEY in ~/.memloom/config.env, not in the client config. The daemon owns every provider call; this server never talks to a provider itself.

Documentation

docs.memloom.dev/mcp/setup

Apache-2.0. Built by Versuno.