The SendSafely MCP Server is currently in limited Beta. Please contact your account manager if you would like to request access. 

The SendSafely MCP (Model Context Protocol) server lets AI assistants interact directly with the SendSafely encrypted file transfer platform — sending packages, managing recipients, uploading/downloading files, administering workspaces, and querying activity logs, all through natural language.

This server works with any MCP-compatible AI client (Claude Desktop, Claude Code, Codex, Cursor, Windsurf, VS Code, and others). The install and credential steps below are identical regardless of which client you use, only Step 4 differs, since each client stores its MCP configuration in a different place.

A more detailed overview of working with the SendSafely MCP Server once you have installed it can be found here. 

Pre-requisites

• Node.js 24.14.1 or newer  (The server is written for Node 24; older versions will not install.)
• npm 11.11.0 or newer (Ships with Node 24+; rarely an issue.)
• A SendSafely account  (Including an API key & secret. How to obtain.)
• The .tgz file (Request a copy of this install file from Support@sendsafely.com. Copy it anywhere — no need to extract.)
• Network access to npmjs.com (The install pulls a few transitive dependencies from the public npm registry.)

Installation Steps:

Step 1. Check your Node version

1a. Open a terminal and run:

node --version

You need v24.14.1 or newer. If you see a lower version or `command not found` then choose from one of the options below to update/install:

macOS / Linux

Recommended: Install nvm , then run:

nvm install 24
nvm use 24

Alternative: Use the official installer: download from nodejs.org and pick the latest "Current" release (24.x).

Windows

Use nvm-windows or the nodejs.org installer.

Then re-check:

node --version    # should print v24.x or higher
npm --version     # should print 11.x or higher

Step 2. Install the server from the tarball

2a. Create a dedicated directory to host the install, then install the tarball into it:

macOS / Linux*

mkdir -p ~/sendsafely-mcp
cd ~/sendsafely-mcp
npm install ~/Downloads/sendsafely-mcp-server-stdio-0.1.0.tgz

Windows*

mkdir $HOME\sendsafely-mcp
cd $HOME\sendsafely-mcp
npm install $HOME\Downloads\sendsafely-mcp-server-stdio-0.1.0.tgz

*Replace `~/Downloads/...` with whatever folder you saved the `.tgz` in. 

The install takes ~15 seconds and pulls in about 170 dependencies from the public npm registry. npm auto-creates a minimal `package.json` for the directory.

2b. Note these two absolute paths — you'll need them both in Step 4:

1. Path to node — run `which node` (macOS/Linux) or `where.exe node` (Windows). 
   • Typical output: `/Users/you/.nvm/versions/node/v24.14.1/bin/node`
2. Path to the installed server — should be exactly:
   • macOS/Linux: `~/sendsafely-mcp/node_modules/@sendsafely/mcp-server-stdio/dist/server.js`  
         (expands to something like `/Users/you/sendsafely-mcp/node_modules/@sendsafely/mcp-server-stdio/dist/server.js`)
   • Windows: `C:\Users\you\sendsafely-mcp\node_modules\@sendsafely\mcp-server-stdio\dist\server.js`

2c. Verify the server file exists:

macOS / Linux

ls ~/sendsafely-mcp/node_modules/@sendsafely/mcp-server-stdio/dist/server.js

Windows

Test-Path $HOME\sendsafely-mcp\node_modules\@sendsafely\mcp-server-stdio\dist\server.js

Step 3. Configure credentials in OS keychain

Credentials are stored encrypted in:

• macOS Keychain (on macOS)
• Windows Credential Manager (on Windows)
• Secret Service (on Linux — requires gnome-keyring or similar)

3a. Run the credentials helper via node, pointing at the installed script. Pick the form matching your OS:

macOS / Linux

node ~/sendsafely-mcp/node_modules/@sendsafely/mcp-server-stdio/dist/setupCredentials.js

Windows

node $HOME\sendsafely-mcp\node_modules\@sendsafely\mcp-server-stdio\dist\setupCredentials.js

3b. when prompted, enter the below:

1. SendSafely URL → paste your portal URL including the `https://` prefix (e.g., `https://yourcompany.sendsafely.com`). 
   • DO NOT enter a bare hostname — credentials save but the first tool call fails with a connection error.
2. API Key → from your SendSafely account settings
3. API Secret → from your SendSafely account settings

3c. Verify what's configured:

node ~/sendsafely-mcp/node_modules/@sendsafely/mcp-server-stdio/dist/setupCredentials.js --status

To remove credentials later:

node ~/sendsafely-mcp/node_modules/@sendsafely/mcp-server-stdio/dist/setupCredentials.js --clear

Step 4 — Connect the server to your AI client

Every client below uses a stdio MCP server pointed at the same two paths from Step 2 — only the config file location (and occasionally the JSON key name) changes.

Claude Desktop

Edit (or create):

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

(Claude Desktop → Settings → Developer → Edit Config also opens this file.)

{
  "mcpServers": {
    "sendsafely": {
      "command": "/Users/you/.nvm/versions/node/v24.14.1/bin/node",
      "args": [
        "/Users/you/sendsafely-mcp/node_modules/@sendsafely/mcp-server-stdio/dist/server.js"
      ]
    }
  }
}

If other servers are already configured, merge sendsafely into the existing mcpServers object rather than replacing it.

Claude Code

Register from the terminal (adjust paths to match yours):

claude mcp add sendsafely -- /Users/you/.nvm/versions/node/v24.14.1/bin/node /Users/you/sendsafely-mcp/node_modules/@sendsafely/mcp-server-stdio/dist/server.js

Or add the same JSON block as Claude Desktop above under mcpServers in .mcp.json (project scope) or your user-level Claude Code config.

Cursor

Edit ~/.cursor/mcp.json (global) or .cursor/mcp.json (project-scoped). Same schema as Claude Desktop:

{
  "mcpServers": {
    "sendsafely": {
      "command": "/Users/you/.nvm/versions/node/v24.14.1/bin/node",
      "args": [
        "/Users/you/sendsafely-mcp/node_modules/@sendsafely/mcp-server-stdio/dist/server.js"
      ]
    }
  }
}

Windsurf

Edit ~/.codeium/windsurf/mcp_config.json — same mcpServers schema as above.

VS Code (MCP / Copilot, Cline, and similar extensions)

VS Code’s native MCP support uses servers (not mcpServers) in .vscode/mcp.json, and typically expects a type field:

{
  "servers": {
    "sendsafely": {
      "type": "stdio",
      "command": "/Users/you/.nvm/versions/node/v24.14.1/bin/node",
      "args": [
        "/Users/you/sendsafely-mcp/node_modules/@sendsafely/mcp-server-stdio/dist/server.js"
      ]
    }
  }
}

Extensions like Cline typically use the same mcpServers schema as Claude Desktop instead — check your extension’s MCP settings panel.

OpenAI Codex (CLI and desktop app)

Codex’s CLI and desktop app (the desktop experience runs inside the ChatGPT app) share a single config file — you only need to edit it once for both surfaces.

Edit (or create) ~/.codex/config.toml and add:

[mcp_servers.sendsafely]
command = "/Users/you/.nvm/versions/node/v24.14.1/bin/node"
args = ["/Users/you/sendsafely-mcp/node_modules/@sendsafely/mcp-server-stdio/dist/server.js"]
startup_timeout_sec = 60

Use the same node and server.js paths from Step 2 — Codex needs the full absolute path to node because the desktop app launches with a minimal PATH and won’t find a bare node.

Load it:

• Terminal (CLI): just run codex — it reads config.toml on launch.
• Desktop app: if it was already open, fully quit and relaunch so it re-reads the config (or run codex app from the terminal).

Verify: prompt Verify my SendSafely credentials are working. — Codex should call verify_credentials and report success.

Other MCP-compatible clients

Any client that supports stdio MCP servers can run this server. You need two things from Step 2: the path to node, and the path to server.js. Consult your client’s MCP documentation for where it stores server config and which JSON key it expects (mcpServers vs servers are the two common conventions).

Important (all clients): the SendSafely URL you entered in Step 3 must include the https:// prefix — this is client-independent.

Step 5. Restart and verify

5a. Fully quit and relaunch your AI Client, just closing the window isn't enough. 

• On macOS, press ⌘Q —
• On Windows, right-click the tray icon → Quit.

5b. Relaunch your Client

5c. Open Settings → Developer → MCP Servers. Look for `sendsafely` — it should show as connected with a green indicator.

5d.  In a new chat, try: `Verify my SendSafely credentials.`

If the connection fails, expand the `sendsafely` entry in the Developer panel for the error message, and re-confirm the node and server.js paths from Step 2.

Using the MCP Server 

Once connected, your AI client has access to several tools covering the receive flow (list received packages, fetch package metadata, decrypt messages, download files), the send flow (create packages, upload files, add recipients, finalize and deliver), workspace management (directories, file rename), and account-level utilities (enterprise info, activity log).

Example prompts: 

• `Verify my SendSafely credentials.`
• `What packages have I received in SendSafely?`
• `Here is a SendSafely link: <link> — please download and decrypt the files.` 
• `Create a SendSafely package, upload /path/to/report.pdf, add alice@example.com as a recipient, and finalize it.`

A more detailed overview of working with the SendSafely MCP Server can be found here.