MCP server
Read your leads and manage your workspace from an AI assistant like Claude Code or claude.ai.
Reply First runs a remote MCP server — a standard endpoint that lets AI assistants (Claude Code, claude.ai, Codex) operate your workspace on your behalf. Ask them to triage leads, audit your setup, or draft knowledge pages, and they call Reply First directly. Everything is scoped to what you're allowed to see and do.
Install
Add the server to Claude Code:
claude mcp add --transport http reply-first https://replyfirst.ae/api/mcpYou can also copy this command from the Copy page menu at the top of any docs page, or add the server from Settings → Integrations → MCP.
Authentication
The endpoint uses OAuth 2.1 — your assistant opens a browser to sign in with your Reply First account, and you approve access. No key is pasted. Prefer a token? Create a personal access token (rf_mcp_…) under Settings → Integrations → MCP and use it as a bearer token. Either way, the assistant acts as you, with exactly your permissions.
What it can do
The server exposes tools across five areas:
| Area | Tools |
|---|---|
| Orgs | List your orgs, get an overview of one. |
| Setup | Read and update behavior settings and the web widget. |
| Knowledge | List, read, create, update, and archive knowledge pages. |
| Analysis | List and read leads, lead and volume stats, performance analytics, handoffs. |
| Copilot | Pause/resume the AI, resolve a handoff, update a qualification, send a manual reply. |
Safety
The same rules that protect you in the dashboard apply here:
Copilot tools act for real. Sending a manual reply sends an actual message to the lead and pauses the AI on that conversation. Assistants confirm the text and target with you first.
Owner-scoped actions (manual replies, pausing AI, resolving handoffs) only work on your own leads, and everything is filtered by your org and view permissions — an assistant can never reach data you couldn't.