# skillmem

> skillmem is an open-source MCP server for knowledge and memory. Local skill memory for Claude Code and Codex: useful skills grow, stale ones fade. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** CLI, macOS, Windows, Linux
- **Website:** https://skillmem.dev
- **Repository:** https://github.com/liza-studio/skillmem
- **MCP Registry name:** io.github.liza-studio/skillmem
- **Categories:** knowledge and memory, developer tools
- **Listing updated:** 2026-11-15

## About

Where most memory tools store facts, skillmem stores procedures: a trigger, the steps taken, the outcome and lessons learned. Everything sits in a plain SQLite file on your disk, and reads and writes make no LLM calls, so there is no API key and no per-use cost.

Skills are ranked by how often they actually helped. Strength rises only on evidence from outside the agent's own judgement, such as a passing test, and drops after a failure. Skills left unused fade on an Ebbinghaus-style schedule, and dead ones are moved to a backed-up archive rather than deleted.

Every memory records its origin (owner, agent, imported or derived), and only the owner can approve one as a rule. Unapproved memories are injected inside a marked block that says they are data, not instructions. Each change is also appended to a SHA-256 hash chain that `skillmem verify` can check for tampering.

## Features

- **Zero-cost reads and writes:** Storing and recalling skills uses no LLM calls, cloud service or API key.
- **Bilingual hybrid search:** Full-text search with stemming for English and Russian, plus an optional local embedding layer for cross-language recall.
- **Earned skill strength:** Strength grows on external evidence, falls after failures and decays when a skill goes unused.
- **Provenance and owner trust:** Each memory records its source, and only the owner can approve a memory as a rule.
- **Tamper-evident history:** Edits are chained with SHA-256 so after-the-fact changes can be detected.
- **Shared across agents:** Claude Code and Codex use one database, and each record notes which agent wrote it.
- **Markdown export:** `export-all` dumps memory to markdown files with YAML frontmatter that re-import to the same records.

## Good for

- Let a coding agent reuse how it solved a task last time
- Share learned procedures between Claude Code and Codex
- Keep an auditable record of what an agent learned and when
- Stop imported or agent-written rules from acting as trusted instructions
- Back up and move an agent's skills as plain markdown

## Install

### Claude Code (Terminal)

```sh
claude mcp add skillmem -- uvx skillmem
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "skillmem": {
      "command": "uvx",
      "args": [
        "skillmem"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "skillmem": {
      "command": "uvx",
      "args": [
        "skillmem"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "skillmem": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "skillmem"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "skillmem": {
      "command": "uvx",
      "args": [
        "skillmem"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "skillmem": {
      "command": "uvx",
      "args": [
        "skillmem"
      ]
    }
  }
}
```

## Tools

- `mem_learn`: Lets an agent record a skill or note into memory; an agent-written rule stays unapproved until the owner trusts it.

## FAQ

### Does it need an API key or cloud account?

No. It runs locally on SQLite and makes no LLM calls to store or retrieve skills.

### How do I install it?

Run pip install 'skillmem[semantic]', then skillmem init --claude-code --codex. The semantic extra is what enables cross-language recall.

### What licence is it under?

Apache-2.0, with the engine published on PyPI.

### Can skills sync across machines?

Not yet. A skillmem cloud for syncing and team sharing has a waitlist; the local engine does not sync.

### Which operating systems are supported?

The README lists macOS, Windows and Linux, with scheduled upkeep via launchd, schtasks and systemd timers or cron.

### Is skillmem free?

Yes. skillmem is open source, so you can use and self-host it for free.

### How do I add skillmem to Claude Code?

Run this command in your terminal: claude mcp add skillmem -- uvx skillmem

### What are the best alternatives to skillmem?

The closest alternatives to skillmem are REA, Codebase Memory, GitHub, Serena MCP: the IDE for your agent and XERJ.

## Alternatives

- [REA](https://ai-powered.com/mcp/rea): REA is an open-source MCP server for developer tools.
- [Codebase Memory](https://ai-powered.com/mcp/codebase-memory): Codebase Memory is an open-source MCP server for knowledge and memory.
- [GitHub](https://ai-powered.com/mcp/github-2): GitHub is an open-source MCP server for developer tools.
- [Serena MCP: the IDE for your agent](https://ai-powered.com/mcp/serena-mcp-the-ide-for-your-agent): Serena MCP: the IDE for your agent is an open-source MCP server for developer tools.
- [XERJ](https://ai-powered.com/mcp/xerj): XERJ is an open-source MCP server for knowledge and memory.
- [Ouroboros](https://ai-powered.com/mcp/ouroboros): Ouroboros is an open-source MCP server for developer tools.
- [Claude Code Ultimate Guide](https://ai-powered.com/mcp/claude-code-ultimate-guide): Claude Code Ultimate Guide is an open-source MCP server for ai models and llm tooling.
- [Nautilus Compass Memory](https://ai-powered.com/mcp/nautilus-compass-memory): Nautilus Compass Memory is an open-source MCP server for knowledge and memory.

---
Source: https://ai-powered.com/mcp/skillmem
