Connected AIs lets an AI assistant you already use, such as ChatGPT or Claude, work with your ReplyZen workspace through the Model Context Protocol (MCP). You can ask it how your agents are set up, what happened to a comment, or why an automation did or did not fire. If you allow it, it can also preview and apply a small set of configuration changes.
The dashboard stays your home for ReplyZen. MCP is a conversational way to inspect the same workspace.
Private beta and how to request access
MCP is in private beta for existing ReplyZen customers. ReplyZen turns it on per workspace; owners can't switch it on themselves.
- Only a workspace owner can connect an assistant and approve what it can do.
- To request access, go to replyzen.ai/mcp/connect while signed in as the owner, or open Settings → Connected AIs and choose Request beta access. You can also email [email protected] with the workspace you want to connect.
- Once your workspace is enrolled, Settings → Connected AIs shows the setup steps below.
Which assistants you can use
ReplyZen MCP is built for the assistants below. All of them are in beta: we are still running live end-to-end checks with each one, so steps and available actions can change.
| Assistant | Status |
|---|---|
| ChatGPT | Beta, live check pending |
| Claude | Beta, live check pending |
| Codex | Early |
| Claude Code | Early |
| Other MCP clients | Not yet verified |
Every assistant uses the same MCP address:
https://queues.replyzen.ai/mcp/
ChatGPT
Developer mode is web-only and currently available on eligible paid and workspace plans. A workspace admin may need to allow it first.
- In ChatGPT on the web, open Settings → Security and login → Developer mode.
- Open ChatGPT Plugins, choose +, create a developer-mode app named ReplyZen, and paste the MCP address above.
- Choose OAuth and Client ID Metadata Document if ChatGPT offers it. Finish ReplyZen sign-in, workspace selection, and permission review.
- Start a new chat, choose + → Developer mode, select ReplyZen, and send your first prompt.
See OpenAI's developer mode guide for current requirements.
Claude
Remote custom connectors are available across Claude plans. Free includes one connector; on Team and Enterprise an organization owner adds it first.
- Open Customize → Connectors (claude.ai/customize/connectors). On Team or Enterprise, an organization owner first opens Organization settings → Connectors.
- Choose + → Add custom connector (on Team or Enterprise: Add → Custom → Web), name it ReplyZen, and paste the MCP address above.
- Choose Add, then Connect. ReplyZen opens so an owner can choose the workspace and review permissions.
- Enable ReplyZen from the + menu in a conversation, then send your first prompt.
See Anthropic's custom connector guide for current requirements.
Codex
codex mcp add replyzen --url https://queues.replyzen.ai/mcp/
codex mcp login replyzen
The login command opens ReplyZen in your browser for workspace approval.
Claude Code
claude mcp add --transport http replyzen https://queues.replyzen.ai/mcp/
Then open Claude Code and run /mcp to review the server and authenticate.
Other MCP clients
Clients that support Streamable HTTP and OAuth Authorization Code with S256 PKCE can try the same address. We haven't verified any other client yet.
A good first prompt: "Use ReplyZen to list our active agents and show which pages are assigned to each one."
What the permissions mean
When you connect, ReplyZen shows a permission screen. You pick one workspace and choose what the assistant can do there. For assistants ReplyZen can't verify, which today includes ChatGPT and Claude because they register themselves when you connect, only the read permission is selected by default. Check that the address shown on that screen matches the assistant you started from.
| Permission | What it allows |
|---|---|
| See your ReplyZen setup (always on) | Agents and their instructions, automations, pages, and change history. It also includes the text of comments and DM conversations the assistant looks up when it explains what happened. |
| Change agent settings | Tone, moderation, links, and other agent settings, and adding to or replacing an agent's instructions. This changes how future comments and DMs are answered. |
| Change automations | Editing triggers and scope, and turning automations on or off. An active automation can send DMs and publish replies to real people. |
| Assign pages to agents | Moving pages between agents or taking a page away from all of them. Assigning can start replies, moderation, and DMs on that page; unassigning stops replies and moderation there. |
Publishing controls (auto-publish, publish mode, confidence threshold), comment deletion, and creating or deleting agents and automations are not available over MCP. Use the dashboard for those.
What the assistant can do
These are the tools ReplyZen offers. An assistant only sees the ones its permissions allow.
Read (always available)
- Verify the connection (
whoami) - List agents and inspect one agent (
list_agents,get_agent) - List connected pages and their assigned agents (
list_pages) - Explain what happened to one comment (
explain_comment) - List automations and inspect one (
list_automations,get_automation) - Explain why automations did or did not fire for a comment or DM conversation (
explain_automations) - Review agent and automation change history (
get_agent_changes,get_automation_changes)
Change (only with the matching permission)
- Update agent settings and agent instructions (
update_agent_settings,update_agent_instructions) - Update an automation, or which agents and pages it covers (
update_automation,update_automation_scope) - Assign or unassign pages (
assign_pages,unassign_pages) - Revert a recorded agent or automation change (
revert_agent_change,revert_automation_change)
How changes are previewed and confirmed
Every change happens in two steps:
- Preview. The assistant asks ReplyZen for a preview. Nothing is written. ReplyZen returns the exact before and after, plus what it will affect, and a short-lived confirmation.
- Confirm. The change is applied only when that exact preview is confirmed. A confirmation works once and expires.
ReplyZen tells your assistant to show you the preview before it confirms. Whether it actually stops to ask you depends on your assistant's tool-approval settings. If your assistant is set to run tools without asking, it can preview and confirm in one turn. Keep approval prompts on for ReplyZen if you want to OK every change yourself, or connect with read-only access.
What revert can and cannot undo
Every change is recorded with who made it. Agent and automation configuration changes can be reverted from the audit trail, as long as nothing else has edited that agent or automation since. A page unassignment can be put back by assigning the page again.
Revert restores configuration only. It cannot:
- withdraw replies that were already published
- recall DMs that were already sent
- remove likes
- reverse hides or other moderation that already happened on the platform
See activity and revoke access
Open Settings → Connected AIs in your workspace.
- The connection ledger lists every connected assistant, who connected it, its permissions, and when it was last active.
- What your AIs did shows recent activity. Previewed means nothing changed; Applied and Reverted changed saved ReplyZen settings. It never shows your assistant conversations.
- Owners can rename or revoke a connection. Revoking disconnects it immediately. Changes it already applied stay in place, and reconnecting starts a fresh permission review.
If MCP is turned off for your workspace, existing connections stop working but still appear in Connected AIs, so an owner can revoke any they no longer want before MCP is turned back on.
Feedback
Tell us what you want your assistant to do next from the support chat in your ReplyZen dashboard, or email [email protected]. Your assistant can't send feedback to ReplyZen for you.