# Memory Fabric MCP server

> Provides a file-first, local-first memory layer for AI coding assistants using Markdown and YAML instead of a vector database.

- Repository: https://github.com/elViRafa/agentic-memory
- Page: https://hostedmcp.org/servers/elvirafa/agentic-memory
- Transport: stdio
- Stars: 1
- License: MIT
- Latest version: 1.4.5

## Install

### Claude Code

```sh
claude mcp add agentic-memory -- uvx memory-fabric
```

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

```json
{
  "mcpServers": {
    "agentic-memory": {
      "command": "uvx",
      "args": [
        "memory-fabric"
      ]
    }
  }
}
```

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

```json
{
  "servers": {
    "agentic-memory": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "memory-fabric"
      ]
    }
  }
}
```

### VS Code CLI

```sh
code --add-mcp '{"name":"agentic-memory","type":"stdio","command":"uvx","args":["memory-fabric"]}'
```

### Claude Desktop (`claude_desktop_config.json`)

```json
{
  "mcpServers": {
    "agentic-memory": {
      "command": "uvx",
      "args": [
        "memory-fabric"
      ]
    }
  }
}
```

### Codex

```sh
codex mcp add agentic-memory -- uvx memory-fabric
```

### Codex config.toml (`~/.codex/config.toml`)

```toml
[mcp_servers.agentic-memory]
command = "uvx"
args = ["memory-fabric"]
```

### Gemini CLI

```sh
gemini mcp add -s user agentic-memory uvx -- memory-fabric
```

### Gemini settings.json (`~/.gemini/settings.json`)

```json
{
  "mcpServers": {
    "agentic-memory": {
      "command": "uvx",
      "args": [
        "memory-fabric"
      ]
    }
  }
}
```

## Tools (17)

- `initialize_memory_fabric_tool`
- `read_combined_context_tool`
- `read_section_tool`
- `keyword_search_tool`
- `write_local_memory_tool`
- `propose_memory_patch_tool`
- `dream_tool`
- `prepare_dream_payload_tool`
- `apply_dream_results_tool`
- `evaluate_memory_fabric_tool`
- `evaluate_dream_quality_tool`
- `write_memory_store_tool`
- `read_memory_store_tool`
- `list_memory_store_tool`
- `delete_memory_store_tool`
- `write_session_journal_tool`
- `write_failure_memory_tool`

## FAQ

### Does Memory Fabric require a vector database?

No, it uses human-readable Markdown with YAML frontmatter instead of a vector database or embeddings.

### Is my data sent to the cloud?

No, Memory Fabric is local-first and does not send data to any cloud, account, or analytics service.

### How does it handle git commits?

It can automatically record every git commit as episodic memory via an optional post-commit hook.