Connect AI assistants like Claude and ChatGPT to Front to triage, analyze, and reply to conversations.
Recently active
Front’s MCP server is here, and the best part is that you don’t need to invent prompts from scratch after your setup! We tested these templates against real Front inboxes and refined them until they produced accurate, verifiable results. Copy the ones below, swap in your {inbox names, dates, and details}, and you're minutes away from an auto-triaged queue, an account briefing, a trend report, or a knowledge-gap audit. 1. Inbox triage: prioritized action listTeam, personal, or both: Team (personal variant in the bonus section)User profile: Support team leads and managers (running it), support agents (consuming the output in Front)When to use:As a scheduled task that runs every morning or at every shift change: the team starts their shift with the highest-priority items already tagged, surfaced in a pinned view, and assigned, so they can action the top items immediately instead of scanning the queue Ad hoc after a volume spike, outage, or holiday backlog, when the queue is too big to eye
Hey, Having MCP and / or API be able to (at least) read the rules details would be a game changer. Other mail services allow to import / export rules as JSON files, which also helps a lot. Concrete example: Having to screenshot your rules 1 by 1 to share it to Claude to find bugs and inconsistencies is a real pain. The bigger the company, the more relevant this would be. Is this something Front could implement?
We use Front MCP (mcp.frontapp.com) from an inbox agent (Pit Bot / grok@xxx.com) Workflow: a teammate @mentions the agent in a Front comment and asks for a fairing-kit mockup. The agent generates a PNG and needs to post it as an internal comment on that same conversation, visible to CS, not to the customer.What we tried: add_comment with conversationId + body. The MCP schema only has those two fields. get_attachment can download comment images, but nothing can upload them.What we expected: add_comment to take an attachments list, same as Core API POST /conversations/{id}/comments which already documents attachments as multipart/form-data, max 25 MB (https://dev.frontapp.com/docs/attachments-1). Teammates already attach images to comments in the Front UI (we do this daily with color swatches).What happened: we can only post text. We have to drop the mockup in a side chat and ask a human to attach it on the thread. That breaks the workflow.Ask: please add an attachments argument on add_c
You can now connect ChatGPT and Codex assistants to the Front MCP server using the official OpenAI plugin, which greatly simplifies setup! To connect via an official directory:A Front admin (or any user authorized to install integrations from the Front App Store) must connect via the directory links above. Then, any user in your company can use the directory links to connect their own accounts.
I was hoping to run some high-level reports through Claude using Front’s Analytics API, but hadn’t realized the Analytics API isn’t exposed in the MCP. Is there a timeline for this?Here is exactly what I was after:Total inbound message volume by inbox (in a specific calendar month) from the Workload report Topic breakdowns by inbox (in a specific calendar month) from the Topics report Average Response Time from Team Performance (for a specific calendar month) Smart QA / Smart CSAT scores (for a specific month)Here is what Claude said: The MCP doesn't expose Front's Analytics module (Workload, Topics, Team Performance, Smart QA, Smart CSAT reports) — those live in a separate reporting API that isn't part of the connector we have access to.
I connected the Front MCP server to Claude Code and it works well for conversations, messages, drafts, comments and tags. Thank you for shipping it. One gap: there is no way to work with message templates. The OAuth app's Resource permissions grid includes a "Message templates" row with Read, Write and Delete, so I ticked it and re-authorized, but the MCP server still exposes no tool that touches templates. The full tool list I get is: search_conversations, read_conversation, move_conversation, assign_conversation, tag_conversation, update_conversation_status, read_message, send_message, list_drafts, create_draft, update_draft, delete_draft, add_comment, get_attachment, list_channels, list_inboxes, list_tags, list_statuses, list_teammates, list_teams, read_contact, search_contacts, read_account, search_accounts, get_my_identity. The server also advertises no MCP resources. For comparison, the Core REST API does support templates: GET /message_templates works with an API token carrying
When Claude creates a reply draft through the Front MCP server, the draft contains only the body text - no quoted thread history and no signature. Front's own Reply button adds both.The Core API already supports this. POST /conversations/{id}/drafts accepts quote_body, signature_id and should_add_default_signature. The MCP create_draft and update_draft tools expose none of them, so an AI-created draft can never match what Front itself produces.Separately, there is no way to create a forward at all: no forward endpoint exists in the Core API, only in the Plugin SDK (Front.forward() / replyOptions.type: "forward"), which requires a plugin running in the Front UI.Impact: I have to re-add the quoted history and my signature by hand on every AI-drafted reply. The manual rework defeats the purpose of the integration - it is faster to skip the connector and hit Reply.Asks: (1) expose quote_body, signature_id and should_add_default_signature in create_draft and update_draft, ideally with quoti
Does the Front MCP connector have the ability to auto-draft templates for bulk sends to a list of contacts?
Good news :) Hopefully v1.1 will be ready soon. When you query open conversations via the API or MCP, you get the global conversation status not the per-inbox view your sidebar shows. This means archived conversations still show as open if they're active in any other inbox, making it impossible to programmatically replicate what you see in Front. For any team building AI agents or automations on top of Front, this is a fundamental blocker.
We’re happy to announce that we’re now officially listed in Claude’s connector directory! This should make it easy for you to connect to the Front MCP server. Prior to being listed on the directory, you had to connect Claude to Front’s MCP server as a custom connector, which involved setting up a developer app in Front and then configuring the connection details in Claude. You can now connect directly from the directory listing by simply logging into Front when prompted. If you have other directories you’d like to see us listed in, please let us know in the comments!
Hi Front community, We successfully tested the MCP Server feature using OAuth and it works great. However, for our production use case we need this feature available for API Tokens as well.Our scenario: we are building a centralized MCP server hosted on our own infrastructure that acts as a proxy to Front's MCP, consumed by multiple support agents through AI assistants (Claude Code). The server needs to authenticate against Front without per-user OAuth flows. The challenge with OAuth: - The refresh token is tied to a specific user's identity - If that person leaves the team or the token is invalidated, the integration breaks - Managing token rotation in a server environment adds significant operational complexity What we need:API Tokens with MCP Server feature enabled, so we can authenticate our server-side MCP proxy with a stable, long-lived credential that isn't tied to a specific user's session.This would be the ideal solution for teams building centralized MCP integrations on
We have an automatic out-of-office reply switched on in Front for any inbound customer service enquiries. When a customer emails in, that auto-reply becomes the most recent message on the thread.When our AI agent then drafts a response, the Front connector replies to the latest message — which is our own out-of-office, not the customer's email. Because it's replying to our own outbound message, the draft has no recipient and no subject line, so it can't be sent.The fix we're asking for: have create_draft reply to the customer's last inbound message (not just the latest message of any kind) — or give us a way to point it at the customer's message directly. Any of those would let the draft inherit the right recipient and subject.
Already have an account? Login
No account yet? Create an account
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.