The Macha × Front Integration: AI Agents for Shared-Inbox Teams

Macha fits the Front model rather than fighting it. Macha is not a help desk and it isn't a Front replacement — it's an AI agent layer that runs on top of the inbox or help desk you already use. Connect Front once, and Macha's agents can read a conversation and its full contact history, draft and post replies, set status, tag for routing, and — because an internal comment can @mention a specific teammate — quietly loop the right person in, all while acting across your other tools like Shopify, Stripe, and Slack.

This is a hands-on look at what the Front connector actually does (it shipped on June 23, 2026), how to set it up, and where it fits next to Front's own rules and AI.

How we wrote this

The screenshots and tool list here come from the real Macha dashboard and the live Front connector, not a marketing page. For Front itself we've stuck to what's plainly true — it's a customer-communication / shared-inbox platform built for collaborative, relationship-driven support, with teammate assignment, internal comments, and @mentions as core concepts — and we'll say clearly where Front already automates things itself.

What the connector gives your agents

Connecting Front adds 15 tools to any Macha agent, each labelled Read or Write, with write tools confirmation-gated.

Grouped by what they do:

• Read the conversation and the contact — fetch a conversation by ID, search conversations by full-text query, look up a contact, and read a conversation's custom fields. The agent builds context before it acts: who's writing, the history, what's already been said.
• Reply and collaborate — post a public reply to the customer, or add an internal comment. The internal comment is the Front-native part: it supports @mentions, so an agent can drop a note and notify the exact teammate who should pick it up — the way your team already works in Front.
• Route and resolve — apply tags, update status (open, archived, and so on), and set custom fields, so conversations land in the right place and close cleanly.
• Know who's who — list tags and custom fields so the agent uses your real taxonomy instead of inventing one.

Because these are discrete tools, you choose what each agent can touch. A routing agent might read, tag, and @mention only; a front-line agent gets reply and status tools too.

Running agents automatically: triggers

The connector also supports webhook triggers, so an agent runs automatically when a conversation is created or a new message arrives. A message comes in, the trigger fires, the agent reads the thread, and it acts — tagging, replying, updating status, or @mentioning a teammate to escalate. You can set a debounce so a quick back-and-forth batches into a single agent run instead of firing on every message, and triggers auto-disable after repeated failures as a safety backstop.

What you can actually automate

The cross-tool action is where this earns its place:

• Order questions, answered from the source. A "where's my order?" message arrives, the agent looks it up in Shopify, and posts the tracking status as a public reply — then tags and archives the conversation.
• Billing questions, triaged with context. The agent checks Stripe, leaves an internal comment with what it found, and @mentions your billing teammate so the handoff is instant and complete.
• Tidy routing on every new conversation. Categorise by topic, tag, set status, and @mention the owning team — your inbox stays organised without anyone triaging by hand.
• Knowledge-grounded replies. Connect your help centre, Notion, or uploaded docs and the agent answers in your wording, citing what it used.
• Escalations that arrive ready. The human gets a comment with the summary, the order, the relevant policy, and what the agent already tried — not just a flag.

A worked example: a billing question, triaged

Here's what "automate" looks like in Front's collaborative model. A customer messages "I was charged twice for order #882." A new message arrives, the trigger fires, and:

1. The agent reads the thread with Get Conversation and identifies a possible duplicate charge.
2. It checks Stripe (a separate connector on the same agent) and finds two captures against the same order.
3. Rather than reply directly, it adds an internal comment summarising what it found — the two charge IDs and the refund link — and @mentions your billing teammate so the right person is notified instantly.
4. It tags the conversation billing for reporting and leaves the status open for the human to finish.

The customer never gets an unreviewed reply on a money question, but your teammate opens a conversation that's already been investigated. That's the shared-inbox version of automation — the agent does the legwork and hands off cleanly, the way your team already collaborates in Front.

Setting it up

Front uses an API token (a JWT), and setup takes a few minutes:

1. In Front, click your initials (top right) → Settings.
2. Go to Developers → API tokens and click Create API token (Front's docs cover this; you must be a company admin). Give it a name and the scopes you need — Shared Resources is enough for read + reply tools.
3. Copy the token — Front shows it only once.
4. In Macha, open Connectors → Support → Front, paste the token, and save. Macha validates it before storing.
5. Add the Front tools to an agent (or start from a built-in template) and switch on a trigger when you're ready.

You can also start from Macha's gallery of pre-built agent templates rather than a blank slate — including Front-specific ones, Front Conversation Triage and Front Reply Assistant:

Prefer not to wire it manually? Macha's AI Agent Builder assembles the agent from a plain-language description — "triage new Front conversations, tag by topic, and @mention the right team."

Where this fits alongside Front's own automation

Straight talk: Front has its own AI, and it's good. Copilot drafts replies from your past conversations and help content for an agent to review and send, and Autopilot learns from those approved replies to handle routine inquiries on its own — Front's whole pitch is a gradual path from assisted to automated. Plus Rules for triage and routing. If everything you need lives inside Front, those native features are solid and you should use them.

One thing worth knowing on cost: Front's Autopilot is a usage-based add-on, billed per resolution — $0.89 each at the time of writing (Front pricing), on top of Professional/Enterprise plans, so the bill scales with volume. Macha instead bills credits per AI action, not per resolution — different model, worth matching to what you're automating (see the pricing page).

Macha isn't competing with Front AI on its home turf — it adds the thing Front AI can't: action across the tools Front doesn't own. Front AI drafts and deflects beautifully inside Front and its help content; it doesn't reach into Stripe to confirm a duplicate charge, push a row to Notion, or hit your internal order API. That cross-tool reach — plus running the same agent design across Front, Zendesk, Freshdesk, and Gorgias — is where Macha sits.

Macha is for what reaches past Front:

• Action across tools Front doesn't own — reading and writing in Shopify, Stripe, Notion, internal APIs via custom tools, and posting to Slack, inside one agent run.
• One agent across every inbox you run — the same Macha agent design works across Front, Zendesk, Freshdesk, and Gorgias, so multi-platform teams don't rebuild automation per tool.
• Configurable, inspectable agents — pick the model, see every tool call and result, gate writes behind confirmation, and test on real conversations first.

It isn't Macha or Front's rules — plenty of teams run Front's native automation for the simple cases and add Macha for the cross-tool, multi-step resolutions.

A note on safety

Writes are gated. Public replies require confirmation in chat; high-impact tools and triggers show an activation advisory before you arm them; and you can keep an agent in internal-comment-only mode so it collaborates with your team — tagging and @mentioning — without ever replying to a customer until you trust it. Test Run lets you watch an agent work a real Front conversation before it touches anything live.

FAQ

Does Macha replace Front? No. Macha is an AI agent layer on top of Front. Front stays your shared inbox; Macha adds agents that read and act across Front and your other tools.

How does Macha connect to Front? With a Front API token from Settings → Developers → API tokens (the Shared Resources scope covers read + reply). The token is shown once; paste it into Macha, which validates and encrypts it.

Can an agent notify a specific teammate? Yes — the Add Internal Comment tool supports @mentions, so an agent can comment and ping the exact person who should take over, exactly as your team does manually.

Can it reply to customers automatically? If you let it. Add Public Reply posts to the customer; you can also keep an agent in internal-comment-only mode. Writes are confirmation-gated by default.

What about Zendesk, Freshdesk, or Gorgias? Same agent model across all of them — one design runs on every help desk Macha supports.

Try it

If your team lives in Front and you want an agent that triages and resolves real conversations, connecting takes a few minutes. Start a 7-day free trial, no credit card required, add Front under Connectors → Support, and point an agent at an inbox — browse all Macha integrations or read the connector docs.