Connect your AI assistant

SendFromChat is a remote MCP server (Model Context Protocol, streamable HTTP). There are two ways to connect:

• Sign in (recommended) — for Claude, ChatGPT and Claude Code. Add https://sendfromchat.com/api/mcp, click Connect, sign in to SendFromChat and click Allow. Nothing secret to copy or paste.
• Private connector URL — for apps without sign-in support (Cursor, VS Code, Windsurf, Gemini CLI, Codex, scripts). The URL carries an API key.

Sign in (Claude, ChatGPT, Claude Code)

The connector URL is the same for everyone:

https://sendfromchat.com/api/mcp

When you connect, your assistant opens SendFromChat's sign-in page. Sign in, check which workspace it's for, and click Allow. That's it — the assistant gets its own access that renews itself, and never sees your password.

Claude (web / desktop): Customize → Connectors → + → Add custom connector → name it SendFromChat → paste the URL → Add → Connect → sign in → Allow.

ChatGPT: Settings → Apps & Connectors (Developer mode on) → create a connector → paste the URL → authentication OAuth → save → sign in → Allow. Enable it in a chat.

Claude Code:

claude mcp add --transport http sendfromchat https://sendfromchat.com/api/mcp

Then run /mcp, choose sendfromchat → Authenticate, and allow it in the browser.

See or disconnect connected apps: Dashboard → Settings → Connected apps. Disconnecting an app stops it immediately.

Private connector URL (other MCP clients)

Many MCP clients now support signing in too, so try https://sendfromchat.com/api/mcp first — if the client opens a browser to connect, you're done. If it doesn't, use a private connector URL instead:

Dashboard → Overview → Connect your AI assistant → Private connector URL → Generate URL. It looks like:

https://sendfromchat.com/api/mcp/wsk_…

The part after /mcp/ is your API key. It's shown once — copy it then. It works like a password; generating a new one disables the old one.

If your client supports custom headers, you can instead use https://sendfromchat.com/api/mcp with Authorization: Bearer wsk_…, which keeps the key out of the URL.

Cursor (~/.cursor/mcp.json):

{ "mcpServers": { "sendfromchat": { "url": "https://sendfromchat.com/api/mcp/wsk_…" } } }

VS Code (.vscode/mcp.json, or Command Palette → MCP: Add Server → HTTP):

{ "servers": { "sendfromchat": { "type": "http", "url": "https://sendfromchat.com/api/mcp/wsk_…" } } }

Windsurf (~/.codeium/windsurf/mcp_config.json):

{ "mcpServers": { "sendfromchat": { "serverUrl": "https://sendfromchat.com/api/mcp/wsk_…" } } }

Gemini CLI (~/.gemini/settings.json):

{ "mcpServers": { "sendfromchat": { "httpUrl": "https://sendfromchat.com/api/mcp/wsk_…" } } }

Codex (~/.codex/config.toml):

[mcp_servers.sendfromchat]
url = "https://sendfromchat.com/api/mcp/wsk_…"

Anything else: add a remote MCP server with the HTTP (streamable HTTP) transport and paste the URL.

First prompt

Say "Help me get set up." The assistant calls get_setup_status and walks you through whatever's left — connecting a number, a first template, a contact list.

How the assistant behaves

• Before messaging anyone it calls check_recipient and picks the cheapest valid option: a reply inside the 24-hour window (free within Meta's monthly service-message limit), an existing approved template, or — only if nothing fits — a new template.
• It previews bulk sends (real rendered messages, how many go out today) and should confirm before sending.
• Everything it does appears in the dashboard.

See the full MCP tools reference.