# agent-memory MCP server

> Provides persistent semantic memory for AI agents with an intelligent decision layer to replay, restore, verify, or skip context injection.

- Repository: https://github.com/TheProdSDE/agent-memory-sdk
- Page: https://hostedmcp.org/servers/theprodsde/agent-memory-sdk
- Transport: stdio
- Stars: 0
- Latest version: 0.1.5

## Install

### Claude Code

```sh
claude mcp add agent-memory-sdk -- uvx agent-memory-sdk
```

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

```json
{
  "mcpServers": {
    "agent-memory-sdk": {
      "command": "uvx",
      "args": [
        "agent-memory-sdk"
      ]
    }
  }
}
```

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

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

### VS Code CLI

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

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

```json
{
  "mcpServers": {
    "agent-memory-sdk": {
      "command": "uvx",
      "args": [
        "agent-memory-sdk"
      ]
    }
  }
}
```

### Codex

```sh
codex mcp add agent-memory-sdk -- uvx agent-memory-sdk
```

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

```toml
[mcp_servers.agent-memory-sdk]
command = "uvx"
args = ["agent-memory-sdk"]
```

### Gemini CLI

```sh
gemini mcp add -s user agent-memory-sdk uvx -- agent-memory-sdk
```

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

```json
{
  "mcpServers": {
    "agent-memory-sdk": {
      "command": "uvx",
      "args": [
        "agent-memory-sdk"
      ]
    }
  }
}
```

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `AGENT_MEMORY_COLLECTION` | no | no | Collection (database) name within the memory directory |
| `AGENT_MEMORY_DIR` | no | no | Directory for the persistent memory store (default: ~/.agent_memory) |

## Tools (7)

- `remember_memory`: Store a query/response pair in persistent agent memory. ttl accepts formats like "30d", "24h", "60m", "3600s".
- `resolve_memory`: Retrieve memory and decide how the agent should respond. Actions: replay, restore, verify, none Modes: auto (default), replay, restore, verify
- `list_memories`: List stored memories with pagination and optional scope filter.
- `get_memory_by_id`: Fetch a single memory by ID.
- `forget_memory`: Delete a memory by ID.
- `archive_memory`: Archive a memory so it is excluded from retrieval.
- `consolidate_memories`: Merge near-duplicate memories into summary entries.

## FAQ

### Can it save money on LLM costs?

Yes, by using the REPLAY action to return stored answers instead of making new LLM calls for repetitive queries.
