Get started with Front's Admin MCP server [open beta]

⚠️ Important to know:

• This feature is in beta. This article may require final changes before the feature is released to general availability (GA).

• Front's Admin MCP server allows external AI tools (e.g., Claude, ChatGPT) to manage your Front settings. To let an AI tool work with your conversations instead, see Get started with Front's MCP server. To let Front's AI connect to an external tool, follow the steps for MCP Connectors.

Front's Admin MCP server lets Front admins connect AI assistants like Claude and ChatGPT to their Front settings. Once connected, the AI assistant can read how your Front instance is set up and create or edit rules, inboxes, tags, message templates, knowledge bases, AI agents, and more, from a request written in plain language.

Every change is made as the admin who connected, only within what their Front role allows, and shows in your audit log.

Use it for configuration work that is slow to do by hand: auditing hundreds of rules, applying the same change across many of them, or setting up and testing an AI agent.

MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external tools. The Admin MCP server gives an AI assistant a set of tools to read and change your Front configuration, the same work you would otherwise do in your settings one item at a time.

It is separate from Front's MCP server. The MCP server works with conversations on behalf of any teammate. The Admin MCP server works with your settings on behalf of an admin.

Think of it as giving your AI assistant access to your Front settings, with your admin role as the limit: it can only change what you could change yourself.

Here are common ways to put the Admin MCP server to work. Copy a prompt, then adapt the names to your own setup.

Audit your rules

"Audit the rules in the Support workspace. List every active rule with its trigger, conditions, and actions. Find exact or near duplicates, and rules that act on the same conversations with conflicting actions, like assigning to two different teammates. Give me the result as a table, worst problems first. Don't change anything."

Merge duplicates into fewer rules

"For each group of duplicates you found, propose one rule that replaces them, or a single branching rule when the rules only differ by one condition. Show me each change before you make it. Once I approve, create the new rule and turn off the rules it replaces."

Understand what a rule does

"Explain in plain language what the VIP escalation rule does, which conversations it applies to, and which other rules can act on the same conversations."

Make the same change across many rules

"Ana is leaving the team. Find every rule that assigns conversations to Ana and list them. Once I confirm, replace Ana with Ben in each one."

Set up an agent from your past conversations

"Look at the last 3 months of resolved conversations in the Support inbox. Group them by what customers asked for, and tell me which requests an AI agent could handle. Then draft the agent's instructions, tone, and a playbook for the top 3 requests. Once I approve, create the agent and the rule that assigns new Support conversations to it."

Test an agent before it goes live

"Create a simulation set for the Returns agent from 20 real resolved conversations in the Support inbox. Cover each of its playbooks and a few requests it should hand off to a teammate. Show me the simulation criteria first, then create the simulations, run the set, and summarize what failed and why."

Find and fix what isn't working

"Review the Returns agent's Opportunities and its heavily edited or discarded drafts from the last 30 days. Group the problems by cause: playbook, instructions, tone, or knowledge. Propose a fix for the top 3. Once I approve a fix, apply it and run the simulation set again to check nothing else broke."

To learn more about simulations and Opportunities, see this article.

If you're building your own internal agent, a company admin can create a private developer app in Front and use its credentials to connect.

Setup involves creating a developer app with the Front Admin MCP feature, choosing its permissions, and pointing your MCP client at mcp.frontapp.com/admin-mcp. You need a separate developer app from the one you use for Front's MCP server. Full instructions are in the Front Developer Portal.

Company admins and workspace admins. Teammates without admin permissions and AI agents can't connect.

• Company admins can work across your whole company.

• Workspace admins can only change the workspaces they administer, and don't see tools for company-level resources such as AI agents, knowledge bases, and custom fields.

For both connection options, admins can limit the read and write permissions the Admin MCP server can use, resource by resource (for example, read-only access to AI agents). These permissions define the full set (superset) of what any connected admin can do through it. From there, each admin is still limited to what their own Front role allows.

• Option 1: Official connector — permissions are controlled by your company's AI assistant administrator (typically the IT or business systems team), depending on the controls the AI assistant offers.

• Option 2: Custom connection — permissions are controlled by the company admin who creates the developer app in Front, as described in the Developer Portal.

Changes made through the Admin MCP server apply to your live Front setup. To stay in control:

• Most AI assistants ask you to approve each edit before it's made. Keep approvals on for write actions.

• Rule edits come back as a draft first. Nothing is saved until you confirm.

• For larger work, ask your AI assistant to check what exists, write a plan, and wait for your approval before changing anything.

• Every change shows in your audit log under the name of the admin who connected.

The Admin MCP server currently lets you:

• Read your setup: inboxes, channels, rules, macros, tags, teammates, teammate groups, views, message templates, signatures, custom fields, shifts, time off responders, Topics, and AI agents

• Create and edit rules, including branching rules and templated rules such as time goal (SLA) and CSAT surveys, and turn rules on or off

• Create and edit inboxes, tags, message templates, views, custom fields, task templates, shifts, and teammate groups, and invite teammates

• Set up AI agents: instructions, tone, Playbooks, knowledge sources, and connectors

• Create, edit, and publish knowledge base articles

• Test AI agents with simulations and simulation criteria, and review Opportunities and AI replies

• Search and read conversations, to learn from your inbox history before you automate it

The Admin MCP server can't delete anything: turn rules off instead, and delete resources in Front. More capabilities are in the works. To see the latest, ask your AI assistant which tools are available once you're connected.

Front's MCP server lets any teammate's AI assistant work with conversations: search, read, reply, comment, tag, and assign. The Admin MCP server lets an admin's AI assistant work with Front settings: rules, inboxes, AI agents, and more. They're separate connections, and you can use both at the same time.

Yes. Tool names, descriptions, and parameters may change before general availability. If a saved prompt or script stops working, ask your AI assistant to refresh its list of tools, or disconnect and reconnect the Admin MCP server.

No. It can create and edit resources and turn rules on or off, but it has no delete tools. AI agents can't be deleted through it either. Delete resources in Front.

In your audit log, under the name of the admin who connected the AI assistant. Only company admins can view the audit log.

Tools depend on your role and plan. Workspace admins don't see tools for company-level resources. Some tools only appear when your plan includes the matching feature: for example, simulation tools need Autopilot, shift tools need Shifts, and connector tools need custom application connectors.

Admin Copilot builds rules, inboxes, and more from inside Front, with nothing to connect. Use the Admin MCP server when you'd rather work from your own AI assistant, or combine your Front setup with other tools connected to it.

Ask your AI assistant to send feedback to Front: the Admin MCP server includes a feedback tool. You can also add an idea in the Front Community.

This feature is free to access during the beta for all plans, but may have restricted usage in the future. Some tools depend on your plan and add-ons: tools for Autopilot agents and simulations require the Autopilot add-on. Rate limits are listed in the Front Developer Portal.