Munin for two-person support teams.

Two people. Maybe four hundred conversations a month. One of you also runs onboarding calls, the other also writes the docs, and neither of you is ever going to fill in a CSAT dashboard. Nobody in this situation is shopping for workforce management. You want the queue to be short when you open it in the morning, and you want the answer one of you typed in March to be findable by the other one in August.

What is the best help desk for a two-person support team?

At two seats the seat price barely matters: Help Scout's free plan covers five users, Chatwoot's free cloud tier covers exactly two, and Crisp's covers two. Choose on two other things — what the AI meter counts once you switch it on, and whether the customer record under the ticket is shared with the rest of your company. Munin is the option where it is.

That is the whole decision, and it is worth being blunt about it, because most comparison tables for small teams are built around a column that has already gone to zero.

What does a help desk cost two people in 2026?

Between nothing and $250 a month, and the spread has almost nothing to do with the number of seats. Three vendors here will take two agents for free. The expensive ones are expensive because of what they charge for the AI, or because they price per ticket instead of per person. Here is the field, with the seat maths done for exactly two people.

Prices were read off each vendor's own page on 25 August 2026 and are indicative: where a vendor shows an annual-billing toggle, the figure above is the annual view, and month-to-month rates are usually higher. Check the quote before you sign.

One correction worth making, because stale listicles still carry it: Groove is gone as a brand. groovehq.com/pricing now redirects to Helply, which charges $1 per ticket with a 250-ticket monthly floor and a $3,000 annual commitment. Any table that still shows Groove at a per-user price is describing a product that no longer sells that way.

Why does the AI meter matter more than the seat price?

Because the seat price is fixed at two and the AI price is not. Two Intercom seats are $58 a month; four hundred conversations resolved by Fin at $0.99 each is $396. Freshdesk's Growth plan is $38 for two agents and includes 500 Freddy sessions, after which sessions cost $0.49 each. The meter you cannot predict is the line item that decides the bill.

The units are also not comparable, which is the part that makes side-by-side pricing pages nearly useless. Help Scout charges per resolution and does not charge when the customer asks for a human afterwards. Intercom charges per outcome, and a qualification is $9.99 rather than $0.99. Front charges per conversation touched by Autopilot, resolved or not. Freshdesk charges per session. Zendesk charges per automated resolution and does not publish the rate anywhere on its own pricing page — which means for Zendesk you cannot do this arithmetic at all until you are on a call with sales. Four vendors, four units, and each of them is counting something other than the work.

Freshdesk has also started metering the agent interface itself. From 1 September 2026 its MCP integration includes 100 actions a month on Growth, 500 on Pro, 1,000 on Enterprise, and $15 per additional thousand — where an action is one successful tool call. That is a meter on the pipe your agent talks through, separate from the meter on the AI. Nobody else in this table has one yet.

Munin does not have an AI meter, because Munin does not sell you any AI. It sells the tools and the record; the intelligence is whatever MCP client you point at it, on whatever subscription you already pay for. The metered number on Munin Cloud Free is tool calls — 5,000 a month — and a supervised two-person desk does not come close to it.

Which of these can two people realistically self-host?

All of Chatwoot, Zammad, FreeScout and Munin, and the honest question is not whether you can but whether you should. Self-hosting adds backups, upgrades and a pager to a team that is already two people deep. If you have an EU residency requirement in a contract, it is worth it. If you are doing it to save $38 a month, it is not.

The licence differences are real and they are frequently misreported. Chatwoot's front page says MIT; its root LICENSE says MIT Expat except the enterprise/ directory, whose own licence states the software "may only be used in production, if you … have agreed to, and are in compliance with, the Chatwoot Subscription Terms of Service." That is a perfectly reasonable open-core arrangement, and Chatwoot is excellent software — it has native WhatsApp and Instagram channels that Munin does not have, and for a team whose customers live on those two apps that single fact should end the comparison in Chatwoot's favour.

Zammad is the cleanest licence in the category: AGPL-3.0 throughout, no ee/ directory, no watered-down community edition, and no licence key needed to self-host the full feature set. FreeScout is AGPL-3.0 as well, and unusually its paid modules are AGPL too — you are buying a download key and support, not different terms. Munin is MIT, which is the most permissive of the four, and it is MIT with nothing carved out; the ability to leave is the argument, not a footnote to it. If you are weighing licences seriously, the longer breakdown of what each open-source project actually licenses you to do is the piece to read next.

What does Munin give a two-person support team?

Six modules on one customer record — conversations, CRM, knowledge base, CMS, outreach, analytics — with 204 MCP tools at one endpoint and no AI billing of its own. Two people get the thing a two-person team cannot otherwise buy: the answer written in a ticket becomes a knowledge-base draft, and the contact it came from is the same row the sales note is on.

That consolidation is the point, and at this size it is worth more than at any other size. A twenty-person support org has someone whose job is to move information between systems. Two people do not. Every integration you would otherwise maintain — helpdesk to CRM, helpdesk to docs, docs to the website — is a job that falls to one of the two of you at 6pm, and the reason those integrations rot is that nobody has time to notice they have.

The agent side is scheduled skills you can read before you trust them. Inbound threads are tagged, titled and matched to a contact before either of you opens the queue. Resolved handovers are proposed as knowledge-base articles, admin-only until a person promotes them. Contacts get enriched from what the customer volunteered rather than from a field one of you meant to fill in. Nothing irreversible happens without approval, and the approvals are the five minutes a day that keep the whole arrangement honest.

• Before you open the queueInbound conversations arrive already tagged with a topic, titled, and resolved to a contact record. The triage pass runs on the conversation.created event, so the sorting is done by the time either of you looks.
• The second person who asksA resolved handover becomes a drafted knowledge-base article. It stays admin-only until one of you promotes it, so the docs grow from real answers rather than from a quarter you set aside to write them.
• The contact recordEnrichment reads what the customer actually volunteered in the thread — name, company, role — and writes it to the CRM row the conversation is already attached to. No second system, no sync.
• Everything irreversibleMerges, sends and publishes are filed as proposals. An agent can propose all day; only a signed-in person approves. That queue is the job the two of you keep.

Who should not pick Munin for this?

Three fairly specific teams, and the criteria are concrete rather than diplomatic.

If your customers reach you on WhatsApp or Instagram, take Chatwoot — it has native channels for both, and Munin's four are email, chat widget, SMS and voice. If you are being measured on SLA timers, CSAT scores or agent-performance dashboards, buy a product that ships them; Munin does not have that reporting layer, and a small desk being audited against contractual response times should not be improvising one. And if you want a mature ticketing UI that a new hire can be dropped into with no explanation, Zammad has spent fifteen years on exactly that and is AGPL all the way down.

What is left after those three is a team that fits Munin unusually well: two people who answer support and own the customer relationship and write the docs, with no third person to move information between those jobs. For them the arithmetic is different from everyone else's. The support tool is also the CRM, so the enrichment happens where the ticket already is. The knowledge base is also the CMS, so the answer that closed a ticket can become a public article without a copy-paste. The AI is a subscription you already hold rather than a per-outcome meter that scales with your busiest month. And because it is MIT with a free EU-hosted tier, the cost of finding out is an afternoon. That combination does not exist at $0 anywhere else in this table, and at two people it is the combination that does the most work.

How should two people actually start?

One channel, ten documents, one agent pass. Wire up email or the chat widget, write the ten questions you are most tired of answering into the knowledge base, and turn on triage only. Give it a fortnight before you add the second pass.

The failure mode is not technical. It is a review queue nobody works. Agents file proposals — merges, drafts, articles — and a queue that goes unread for three weeks becomes a queue that has quietly stopped mattering. Two people can absolutely keep that queue clear; two people who each assume the other is doing it cannot. Decide which of you owns it on day one.

Frequently asked questions

What is the cheapest help desk for two agents? Several are free at two seats. Chatwoot Cloud's Hacker plan covers exactly 2 agents at $0 with a 500-conversation monthly cap, Crisp's Free Forever plan covers 2 agents, and Help Scout's free plan covers up to 5 users on one inbox. Munin Cloud Free is €0 per organisation and includes CRM, knowledge base, CMS, outreach and analytics alongside support.

Do I need a shared inbox or a real help desk at two people? A shared inbox is enough until you need three things: a record of who the sender is beyond their email address, answers that survive being written once, and a way to hand a thread over without forwarding it. When you want those, you want a help desk — and at this size the one to want is the one that also holds the customer record.

Which help desks have an official MCP server in 2026? Help Scout (GA 28 July 2026, read-only), Front (open beta, read and write), Intercom (mcp.intercom.com), Crisp, Freshdesk (early access), and Munin. Zendesk publishes an MCP client rather than a server; the "MCP Server" app in its marketplace is built by a third party. Chatwoot, Zammad and FreeScout have no official server — the ones you find on GitHub are community projects.

How much does AI support cost per ticket? It depends entirely on the unit, and the units are not comparable. Help Scout charges $0.75 per resolution, Intercom from $0.99 per outcome and $9.99 per qualification, Freshdesk $0.49 per Freddy session after 500, Front from $0.05 per Autopilot conversation, Zammad €0.03 per AI call. Zendesk bills per automated resolution but does not publish the rate. Munin does not meter AI at all — you connect your own MCP client and pay your model provider directly.

Is Munin free to self-host? Yes, and permanently. The repository is MIT with no ee/ or enterprise/ directory, so there is no separately-licensed tier and no feature gate; docker compose up brings up the same code that runs on Munin Cloud. You pay in hosting and maintenance instead of licence fees.

Can two people run this without a developer? On Munin Cloud, yes. You need to be comfortable connecting an MCP client such as Claude or ChatGPT and reading a markdown skill to see what a scheduled pass will do before you enable it. Self-hosting is the part that wants someone comfortable with Docker and Postgres.

Two people and an afternoon is enough to find out whether the shape fits: start on Munin Cloud or read the agent recipes first.

Two people don't need a bigger help desk. They need a shorter queue.