# The MCP server (/en/docs/memo/mcp)

This page helps you connect Memo to an AI app that has no terminal, such as Claude Desktop or Cursor.
MCP (Model Context Protocol) is the way these apps use a program on your computer. Once connected,
the app can save memories, record facts, propose actions and read the agent log. What it saves waits
for your review.

## What the app can do through MCP [#what-the-app-can-do-through-mcp]

| The app can | The app cannot |
| --- | --- |
| Add, list, show and search memories | Confirm a memory |
| Add, list and show facts | Accept a fact |
| Propose actions and list them | Approve or reject an action |
| Read the agent log | Change your notes, mail or messages |
| Look up projects, tags and topics by name, and see which tasks are ready | Create or link tasks |

Search, a person's context and notes are not part of Memo's MCP server. An app reaches them through
tg's or max's MCP server (see [the tg MCP server](/llms.mdx/docs/tg/mcp/content.md)), or an agent
with a terminal runs `memo` with `--json`.

## Connect an app [#connect-an-app]

Add Memo to the app's MCP settings. Most apps use a file like this:

```json
{ "mcpServers": { "memo": { "command": "memo", "args": ["mcp"] } } }
```

Add the `memo` entry next to any servers already in the file:

| App | Settings file |
| --- | --- |
| Cursor | `~/.cursor/mcp.json` |
| Claude Desktop, macOS | `~/Library/Application Support/Claude/claude_desktop_config.json` |
| Claude Desktop, Windows | `%APPDATA%\Claude\claude_desktop_config.json` |

Restart the app after you save the file. Memo must be installed on the same computer as the app. If
the app cannot find `memo`, replace `"memo"` with its full path, which `which memo` (macOS and Linux)
or `where.exe memo` (Windows) prints.

Then try it:

> ```text prompt
> Use the memo tools to list the memories waiting for my review.
> ```

## The tools [#the-tools]

| Tool | What it does |
| --- | --- |
| `memories_add`, `memories_list`, `memories_show`, `memories_search` | Save and read memories |
| `facts_add`, `facts_list`, `facts_show` | Record and read facts |
| `actions_propose`, `actions_list`, `actions_show` | Propose actions and read them |
| `agents_log` | Read the agent log |
| `vocabulary_list`, `vocabulary_resolve`, `vocabulary_preflight` | Look up projects, tags and topics, read only |
| `task_relations`, `task_readiness`, `tasks_ready` | See how tasks depend on each other, read only |

Everything the app adds is saved under the agent name `memo-mcp` and waits as proposed. You review it with
`memo memories list --status proposed` and the commands in [Memories and facts](/llms.mdx/docs/memo/memories/content.md).

## The agent log [#the-agent-log]

Every call the app makes is written to the agent log:

```sh
memo agents log --agent memo-mcp --limit 20
```

The log shows the tool, the result and the time, never what the app sent.

Next: [review what the app saved](/llms.mdx/docs/memo/memories/content.md).