Lorekeeper

Agent Long-Term Memory Bank using SQLite and FTS5.

Lorekeeper is a Rust MCP (Model Context Protocol) server that provides structured long-term memory for AI coding agents (operating under the TARS protocol or similar workflows). It replaces flat-file history with a queryable SQLite database, enabling agents to store, search, and retrieve typed memory entries via MCP tools over stdio.

Key Features

• Context Window Management: Reduces per-session token load by allowing selective retrieval.
• Typed Entries: 11 semantic entry types (e.g., DECISION, COMMIT, PLAN, LESSON, SESSION_SUMMARY).
• Role Enforcement: Mechanically prevents unauthorized writes (e.g., Builder agents cannot assert architectural constraints).
• Full-Text Search: Backed by SQLite FTS5 across titles, bodies, and tags.
• Rich Interaction: 11 MCP tools covering CRUD, search, and memory analytics.
• Isolated Storage: Automatically manages a project-local SQLite database (.lorekeeper/memory.db).

Installation

From Source

Requires Rust 1.94.0+ (Edition 2024).

# Clone the repository
cargo install --path .
# Alternatively, if just is installed:
just install

MCP Configuration

To use Lorekeeper with MCP-compatible clients, add the server to your configuration. Since cargo install places the binary in ~/.cargo/bin/, which should be on your PATH, you can usually use the bare command. If it fails to spawn, use the absolute path to lorekeeper.exe.

Note: Set LOREKEEPER_ROOT to the project directory if the server isn't launched from within a project that contains a .git/ or .lorekeeper/ directory.

Antigravity (mcp_config.json):

"lorekeeper": {
  "command": "lorekeeper",
  "args": [],
  "env": {
    "LOREKEEPER_ROOT": "/path/to/your/project"
  }
}

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "lorekeeper": {
      "command": "lorekeeper",
      "env": {
        "LOREKEEPER_ROOT": "/path/to/your/project"
      }
    }
  }
}

Usage for Agents

Lorekeeper provides the following MCP tools for agentic workflows:

• Write: lorekeeper_store, lorekeeper_update, lorekeeper_delete
• Read: lorekeeper_get, lorekeeper_search, lorekeeper_recent, lorekeeper_by_type, lorekeeper_render
• Health: lorekeeper_reflect (detects duplicates/staleness)
• Meta: lorekeeper_stats, lorekeeper_help

Agents can self-discover capabilities by calling lorekeeper_help.

Development

The project uses just for automation.

# Run the verification pipeline (fmt, clippy, test)
just check

# Run tests only
just test

# Build the project
just build

The database resides locally at <PROJECT_ROOT>/.lorekeeper/memory.db.