> ## Documentation Index
> Fetch the complete documentation index at: https://docs.plugkit.co/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP server

> Give Claude, Cursor, VS Code or any MCP client the ability to publish, read and reply on your social accounts.

PlugKit ships a hosted **MCP server**: the same capabilities as the REST API,
exposed as tools an AI assistant can call. Point a client at one URL, sign in,
and the assistant can list your connected accounts, publish a post, triage the
inbox, or pull analytics — no code, no SDK.

```
https://api.plugkit.co/mcp
```

<Note>
  Transport is **Streamable HTTP** (stateless). There is nothing to install and
  nothing to run locally — no `npx`, no local proxy.
</Note>

## Connect in one click

<CardGroup cols={2}>
  <Card title="Claude" icon="message" href="/mcp/clients#claude-web-desktop-mobile">
    Settings → Connectors → Add custom connector → paste the URL.
  </Card>

  <Card title="Claude Code" icon="terminal" href="/mcp/clients#claude-code">
    `claude mcp add -t http plugkit https://api.plugkit.co/mcp`
  </Card>

  <Card title="Cursor" icon="arrow-pointer" href="/mcp/clients#cursor">
    One-click install link, or three lines in `~/.cursor/mcp.json`.
  </Card>

  <Card title="VS Code" icon="code" href="/mcp/clients#vs-code">
    One-click install link, or `code --add-mcp`.
  </Card>
</CardGroup>

Fastest path of all: the **[MCP section of the dashboard](https://plugkit.co/dashboard/mcp)**
has an install button per client, a ready-to-paste config for everything else,
and the list of assistants you've already connected.

## Authentication

Two ways in, same account scoping as the REST API — an assistant only ever sees
the profiles, accounts, posts and conversations of the PlugKit user it acts for.

<Tabs>
  <Tab title="Sign in (OAuth)">
    The recommended path, and the only one that needs no secret in a config file.
    PlugKit is a full OAuth 2.1 authorization server (PKCE, no client secret): the
    client sends you to a PlugKit consent screen, you approve, and it stores a
    token of its own.

    You choose what you grant:

    | Scope       | What the assistant can do                                                                          |
    | ----------- | -------------------------------------------------------------------------------------------------- |
    | `mcp:read`  | Read profiles, accounts, posts, conversations and analytics. Write tools are not even listed.      |
    | `mcp:write` | Everything above, plus publish, send messages, reply to comments, connect and disconnect accounts. |

    Revoke any assistant at any time from **Settings → Connected apps** in the
    dashboard.
  </Tab>

  <Tab title="API key">
    For clients without OAuth support, send a key as a plain Bearer header:

    ```json theme={null}
    {
      "mcpServers": {
        "plugkit": {
          "type": "http",
          "url": "https://api.plugkit.co/mcp",
          "headers": { "Authorization": "Bearer sk_your_key" }
        }
      }
    }
    ```

    Create the key in the dashboard's [API Keys](https://plugkit.co/dashboard/api-keys)
    section. A key carries **full access** — it has no scopes — so prefer OAuth
    when the client supports it.
  </Tab>
</Tabs>

## What an assistant can do

<CardGroup cols={2}>
  <Card title="Publish" icon="paper-plane">
    Draft, schedule and publish to several accounts at once, with media.
  </Card>

  <Card title="Inbox" icon="inbox">
    Read and answer Instagram, WhatsApp and Telegram conversations.
  </Card>

  <Card title="Comments" icon="comments">
    Read comments on a published post and reply publicly.
  </Card>

  <Card title="Analytics" icon="chart-line">
    Followers, reach, and per-post metrics where the platform exposes them.
  </Card>

  <Card title="Connections" icon="plug">
    Hand the user an OAuth link to connect a new account, or disconnect one.
  </Card>

  <Card title="All 37 tools" icon="wrench" href="/mcp/tools">
    The full reference, plus the three ready-made workflows.
  </Card>
</CardGroup>

## Safety

Publishing a post, sending a message and replying to a comment are **public and
immediate**. The server instructs the model to show the exact content and get
your approval before calling those tools, but that is a prompt, not a lock — the
real guarantees are the ones you control:

* Grant `mcp:read` only, when you just want reporting and triage.
* Review the accounts a connected assistant can reach with `whoami`.
* Revoke from **Settings → Connected apps** the moment you're done.

<Warning>
  `delete_profile` cascades to its accounts, posts and conversations, and
  `revoke_api_key` can revoke the very key the connection runs on. Both are
  write-scope tools — another reason to hand out `mcp:read` by default.
</Warning>