# Leteo [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/asanabrial/leteo  
**GitHub Stars:** 3  
**npm Downloads (last month):** 815  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/leteo

## Description
Persistent memory for AI coding agents: one Rust binary, one local SQLite database.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `bunx` (confidence: high):

```json
"mcpServers": {
  "leteo": {
    "command": "bunx",
    "args": ["@asanabrial/leteo"]
  }
}
```

## Documentation

## What Leteo MCP server does

Leteo MCP server adds persistent memory to AI coding agents. It records information such as implementation decisions, resolved bugs, project conventions, and non-obvious findings in a SQLite database stored on the local machine. The agent can use that memory in later sessions instead of rediscovering the same project details.

The database belongs to the user and can be inspected with Leteo's terminal interface. `leteo tui` opens a dashboard for browsing and narrowing memories, while `leteo export` provides a way to take the stored data elsewhere. `leteo delete` removes it. Cloud replication can be enabled for a project, but the default design is local storage.

## How it works

Leteo runs as one compiled Rust executable. SQLite is included in the binary, so a separate database server, SQLite installation, runtime, or API key is not required. The MCP connection is configured for an agent through Leteo's setup flow, and the agent can save and retrieve memories during its work.

The intended workflow is implicit rather than prompt-driven: an agent begins a session with relevant project knowledge and records useful discoveries as it works. The stored notes are written for future agent sessions and are kept separate from the normal conversation context.

Leteo MCP server is designed for local operation on Linux, macOS, and Windows. The repository provides prebuilt binaries for x86-64 Linux, Windows, and macOS, plus arm64 Linux and macOS builds. The documented Windows target is Windows 10 or later; the Linux releases require glibc 2.34 or newer.

## Setup and configuration

Installers are available through shell and PowerShell scripts, Homebrew, Scoop, winget, and npm. Source installation is also available with Cargo. Release archives include the SQLite-enabled executable, so most installation paths do not require Rust. Building from source requires Rust and Cargo.

After installation, run `leteo tui` and use the Setup screen to configure an agent. A source build can rerun setup for agents that already have Leteo configured, because the MCP entry must point to the new executable path.

The installer scripts accept `LETEO_INSTALL_DIR`, `LETEO_VERSION`, and `LETEO_BASE_URL`. These variables control installation behavior rather than the running MCP process. A source checkout can use `LETEO_SETUP_AGENTS` to provide a space-separated list of agents for setup.

## Tools and capabilities

- Provides 22 MCP tools for agent-facing memory operations.
- Persists memories in a local SQLite database.
- Supports project-specific knowledge across agent sessions.
- Includes a terminal UI for browsing and searching stored memories.
- Exports or deletes the local memory store.
- Supports optional cloud replication for a named project.

## Limitations and notes

Leteo is focused on persistent memory for coding agents, not general-purpose database administration.

The repository's measurement notes report mixed token results and explicitly caution that Leteo does not reliably reduce token use. Its stated purpose is to preserve project knowledge and reduce confidently incorrect answers when the code alone does not contain the needed context. The npm installation route cannot configure an agent through `leteo setup`; the README directs users to configure the agent with the npx line or use another installation method.

_Full upstream README: https://allmcps.com/mcp/leteo/readme_