---
title: "Sleevy MCP Server"
description: "Connect an AI agent to save, find, and organize your Sleevy reading queue."
canonical: "https://sleevy.app/docs/mcp"
source: "https://sleevy.app/docs/mcp.md"
---

# Sleevy MCP Server



Sleevy supports the [Model Context Protocol](https://modelcontextprotocol.io/) (MCP). Connect it to an AI agent to save links, search their Readable Content, organize folders, and manage your reading queue without leaving your workflow.

## Install Sleevy [#install-sleevy]

Run this command in a terminal:

```bash
npx add-mcp https://api.sleevy.app/mcp --name sleevy
```

The installer detects supported MCP clients and lets you choose where to add Sleevy. It uses the Streamable HTTP endpoint at `https://api.sleevy.app/mcp`.

Machine-readable discovery is available through the [Sleevy AI Catalog](https://sleevy.app/.well-known/ai-catalog.json), which points clients to the [Sleevy MCP Server Card](https://api.sleevy.app/mcp/server-card).

## Sign in and permissions [#sign-in-and-permissions]

The first time your agent connects, Sleevy opens an OAuth sign-in and consent flow. Sign in, then approve only the permissions your agent needs. You do not need to add a personal API key to the MCP configuration.

<Callout title="Keep confirmations enabled">
  Removing a saved item or folder is permanent. Keep your agent's MCP tool-confirmation prompts enabled for destructive actions.
</Callout>

## Available tools [#available-tools]

| Capability           | Tools                                                |
| -------------------- | ---------------------------------------------------- |
| Save and list links  | `save_link`, `list_saved_items`                      |
| Search saved reading | `search_saved_content`                               |
| Organize saved items | `set_saved_item_read_state`, `set_saved_item_folder` |
| Manage folders       | `list_folders`, `add_folder`                         |
| Remove content       | `delete_saved_item`, `remove_folder`                 |

The tools available to an agent are determined by the permissions granted during OAuth consent. Reconnect Sleevy to review or change those permissions.

`search_saved_content` accepts a natural-language question or search phrase and returns the most relevant passages with their section and original URL. It combines keyword and semantic search over Readable Content. The agent should base its answer on those excerpts, cite the URLs, and say when the saved material does not provide enough evidence. It uses the existing `saved-items:read` permission.

### Looking before you connect [#looking-before-you-connect]

Describing the server takes no credential. A client may complete `initialize`
and read `tools/list` against `https://api.sleevy.app/mcp` without signing in,
and it gets the full tool list back — every tool, not only the ones some scope
would unlock — so it can decide whether connecting is worth sending you to a
consent screen at all.

```bash
curl -s https://api.sleevy.app/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
```

Calling a tool is a different matter. Every `tools/call` needs a credential and
the scope that covers it, and an unauthenticated one is refused with `401` and a
`WWW-Authenticate` header pointing at the protected-resource metadata. A batch
is allowed only if every message in it is, so a `tools/call` cannot ride along
beside an `initialize`.

The same list is also available without opening a connection at all, as the
[MCP Server Card](https://sleevy.app/.well-known/mcp/server-card.json).

## Connect manually [#connect-manually]

If your MCP client has a manual connection screen, use this server URL:

```text
https://api.sleevy.app/mcp
```

Choose the client’s HTTP or Streamable HTTP transport option, then complete the OAuth sign-in flow when prompted.