Connecting Claude to FreeScout

Two options, explained step by step – no technical background needed

With McpConnect, Claude, the AI assistant from Anthropic, works together with your FreeScout. There are two options for this. They don't rule each other out, and many teams use both. This guide first explains the difference and then walks you through each option separately.

The two options at a glance

Option 1 – Claude as your personal assistant

You work as usual in Claude Desktop or on claude.ai and connect Claude to FreeScout there. Then you ask Claude in the chat, for example:

• “Which requests came into the Support mailbox today?”
• “Summarize ticket #1234 for me in three sentences.”
• “Write a draft reply for ticket #1234.”

Claude works only when you ask it to, and it does so on your behalf with your FreeScout permissions. Each person connects their own Claude.

Option 2 – Claude as an agent in FreeScout

FreeScout gets its own AI colleague, for example “James”. He works on his own in the background: he reads every new customer message, summarizes it in a note, prepares replies, assigns tickets according to your rules and – if you allow it – can send safe replies himself, including in the live chat. On top of that, he handles tasks from your team in the AI (KI) menu.

The agent works under its own FreeScout user and connects through Anthropic's programming interface (API). An administrator sets it up once for everyone.

Comparison at a glance

Pros and cons

Option 1 – Personal assistant

• ✅ quick to set up, no extra cost on top of your existing Claude plan
• ✅ you decide yourself, with every question, what Claude does
• ✅ ideal for reports, summaries and drafts on demand
• ✅ Claude can combine FreeScout with other connectors, such as your calendar or your documents
• ❌ only works when someone asks – new tickets sit there until you point Claude at them
• ❌ each person has to connect Claude themselves
• ❌ can't reply to customers on its own and can't help in the live chat

Option 2 – Agent in FreeScout

• ✅ works on its own: every new request is sorted and prepared before anyone opens it
• ✅ answers safe questions itself, including in the live chat, if you allow it
• ✅ one setup for the whole team, one place for rules, budget and emergency stop
• ✅ recurring tasks, such as a weekly report, run by themselves
• ❌ its own pay-as-you-go cost at Anthropic (can be capped with a daily budget)
• ❌ more setup and a few days of dry run until the rules are right
• ❌ needs its own FreeScout user

Which option suits you?

Option 1 suits you if you are a small team, everyone already has a Claude plan and you don't want any automated workflows. Claude then helps on demand, and each person stays in control.

Option 2 suits you if not everyone has a Claude plan or automation is the main goal: new requests should be sorted, prepared or answered automatically, including in the live chat. All it takes is a single Anthropic account for the whole company, and everyone on the team benefits – even without their own Claude plan: through a note to the agent and, if you enable it for everyone, through the AI menu.

How this guide is organized

Not an administrator and only want to use Option 1? Then ask your administrator whether Part A and step B1 are done, get the MCP server address from them and start at B2.

A few terms first

You don't need to know these terms to follow the guide. But they help if you come across them along the way.

Part A – Preparing FreeScout (both options)

Someone with administrator rights in FreeScout does this part once.

A1 – Opening the AI connections area

1. Sign in to FreeScout as an administrator.
2. In the blue bar at the top, click Manage (Verwalten).
3. In the list, click AI connections (KI-Verbindungen).

A column with the McpConnect areas now appears on the left: Clients, Connections (Verbindungen), Rights (Rechte), Agent, Knowledge (Wissen) and more.

A2 – Running the self-check

The self-check makes sure your FreeScout is ready, for example whether HTTPS is set up and whether sign-in is forwarded correctly.

1. On the left, click Self-check (Selbstprüfung).
2. Click Check now (Jetzt prüfen).
3. Wait a few seconds until the list appears.

How to read the result:

A3 – (Optional) Checking the rights

Under Rights you can see the most an AI is ever allowed to do in your FreeScout. This applies to both options, no matter what anyone ticks later on. By default:

• allowed directly: reading, internal notes, reply drafts, changing the status
• only after approval by a person: marking as spam, moving to the trash, moving to another mailbox, changing customer details
• not possible at all: deleting permanently, adding new recipients to replies, changing customer email addresses

You don't need to change anything here to get started. ✅

Part B – Option 1: Claude Desktop and claude.ai

What you need for Option 1

• Part A is done (administrator).
• A FreeScout user account with an email address and password.
• A Claude account. All plans work: Free, Pro, Max, Team and Enterprise. The free plan (Free) allows exactly one custom connector.
• The MCP server address, for example https://support.deine-firma.ch/mcp (see B1).
• An up-to-date browser, for example Chrome, Edge, Firefox or Safari.

B1 – Allowing Claude and passing on the address (administrator, once)

Allowing Claude:

1. Under Manage → AI connections, click Settings (Einstellungen) on the left.
2. Find the field Client ID metadata documents (Client-ID-Metadaten-Dokumente) and enter: claude.ai
3. Just above it, under Who may connect AI clients (Wer KI-Clients verbinden darf), check that the right roles are ticked: “Users” (Benutzer), “Administrators” (Administratoren) or both.
4. Scroll all the way down and click Save (Speichern).

Passing on the address:

1. On the left, click Clients.
2. At the very top you'll see MCP server address for the client (Adresse des MCP-Servers für den Client) with an address ending in /mcp.
3. Select the address and copy it (Mac: ⌘ + C, Windows: Ctrl + C).
4. Pass it on to everyone who wants to connect Claude, for example by email.

B2 – Adding the connector in Claude (Free, Pro, Max)

Do you have a Team or Enterprise account through your company? Then continue with B3.

1. Open claude.ai in your browser and sign in.
2. On the left, click Customize (Anpassungen), then the Connectors (Konnektoren) tab at the top.

1. On the right, click Add (Hinzufügen) and then Add custom connector (Benutzerdefinierten Connector hinzufügen).

1. Fill in the window:
   • Name: for example FreeScout. This is the name you'll see in Claude later.
   • Remote MCP server URL (MCP-Server-URL): the address from B1, for example https://support.deine-firma.ch/mcp. Make sure it starts with https:// and ends with /mcp, with no spaces.
2. Click Continue (Weiter).

1. Claude checks the address and shows the sign-in settings. They are already preselected correctly:
   • under «Authentifizierung» (authentication): «Jetzt anmelden» (sign in now) – with the note «Erkannt» (detected)
   • under «OAuth-Client» (OAuth client): Use Claude's published identity (Claudes veröffentlichte Identität verwenden) – with the note «Empfohlen» (recommended)
2. Don't change anything and click Add (Hinzufügen).
3. Claude now shows the page of the new connector with the note «Du bist noch nicht mit FreeScout verbunden» (you're not connected to FreeScout yet). Click Connect (Verbinden) there.

Continue with B4.

B3 – Connecting the connector (Team, Enterprise)

With Team and Enterprise accounts, a person with owner rights (Owner) adds the connector once for the whole organization:

1. Owner: In «Organisationseinstellungen» (organization settings), open the Connectors (Konnektoren) area → Add (Hinzufügen) → «Benutzerdefiniert» (custom) → choose «Web».
2. Owner: Enter the name FreeScout and the address from B1 → Continue (Weiter) → keep the preselected sign-in settings (Use Claude's published identity) → Add (Hinzufügen).

After that, each person connects on their own:

1. Open Customize (Anpassungen) → Connectors (Konnektoren).
2. Find FreeScout in the list. The entry usually carries the label «Benutzerdefiniert» (custom).
3. Click Connect (Verbinden).

B4 – Signing in to FreeScout

After you click, Claude opens a new window or tab with your FreeScout.

• If you're already signed in to FreeScout in this browser, you go straight on to B5.
• Otherwise, the FreeScout sign-in page appears. Enter your usual FreeScout email address and password there and click Login (Anmelden).

B5 – Deciding in the consent dialog what Claude may do

Now comes the most important step. FreeScout shows the page Allow Claude to access FreeScout? (Darf Claude auf FreeScout zugreifen?). This is where you decide what Claude is allowed to do. Take a moment for it.

Go through the page from top to bottom:

1. Check the header. It says “Identity confirmed by claude.ai” (Identität bestätigt durch claude.ai) and “Returns to: claude.ai” (Rückkehr zu: claude.ai). Does it say something else, or didn't you start the connection yourself? Then click Cancel (Abbrechen).

2. Tick the permissions.

• Under Read (Lesen), the boxes are already ticked. This lets Claude read and search your mailboxes and conversations. Claude needs this for almost every question.
• Everything that writes is marked with an orange badge writes (schreibt) and is never preselected. Only tick what you really need:
  • Add internal notes in your name (Interne Notizen in Ihrem Namen anlegen): Claude may write notes in tickets. Customers don't see them.
  • Create reply drafts in your name (Antwortentwürfe in Ihrem Namen anlegen): Claude may prepare replies. They are never sent automatically; you check and send them yourself.
• Depending on the setup, you'll see more rows, for example for the status of tickets or for other modules. Read each row before you tick a box.

3. Choose mailboxes. The default is “All mailboxes I can see, including future ones” (Alle Postfächer, die ich sehe, auch künftige). If you only want to show Claude certain mailboxes, choose “Only these mailboxes:” (Nur diese Postfächer:) and tick them.

4. Personal data. If Pseudonymize (Pseudonymisieren) is ticked, Claude sees placeholders such as “Vorname.1” instead of real names, addresses, phone numbers and email addresses. Leave it ticked if you're unsure. If your administrator has switched pseudonymization on permanently, you can't untick it here.

5. Allowed IP addresses – leave empty. This field is meant for special cases.

6. Allow access. At the very bottom, click Allow access (Zugriff erlauben). The window closes or takes you back to Claude.

B6 – Checking that it worked

1. Claude briefly shows «Mit FreeScout verbunden» (connected to FreeScout). On the connector's page, the top right now shows Disconnect (Trennen) instead of Connect, and below it the list of tools Claude may use in FreeScout.
2. You'll receive an email from FreeScout with the subject “New AI connection to your FreeScout account” (Neue KI-Verbindung zu Ihrem FreeScout-Konto). This is normal and a security measure: if you didn't make the connection yourself, revoke it immediately (see Part D).

The tool permissions. For each tool there are three small icons on the right: the check mark for «Immer erlauben» (always allow), the hand for «Genehmigung erforderlich» (approval required) and the crossed-out circle for «Blockiert» (blocked). With the selector next to a group, such as «Schreibgeschützte Tools» (read-only tools, i.e. the ones that only read), you switch all tools in that group at once. To start with, leave everything as it is.

B7 – Using Claude day to day

Turning on FreeScout in the chat:

1. Start a new chat in Claude.
2. Click + at the bottom left of the input field.
3. Hover over Connectors (Konnektoren). A list opens next to it.
4. Turn on FreeScout there (the switch turns blue).

Asking questions. Just write normally, for example:

• “Show me the open tickets in the Support mailbox.”
• “What does the customer want in ticket #1234? Give me a short summary.”
• “Find all tickets from last week that are about invoices.”
• “Write a draft reply for ticket #1234: friendly, short, thanking them for their patience.”

Claude asks for permission. Before Claude looks something up or writes something in FreeScout, it asks you, for example «Claude möchte List mailboxes von FreeScout verwenden» (Claude wants to use List mailboxes from FreeScout). The tool names are in English: “List mailboxes” lists the mailboxes, “Search conversations” searches conversations, “Create reply draft” creates a reply draft. Below that are three buttons:

• «Einmal erlauben» (allow once): only for this one step.
• «Immer erlauben» (always allow): Claude no longer asks for this tool. You can change this back at any time in the tool permissions (B6).
• «Ablehnen» (deny): Claude may not carry out this step.

Once you allow it, Claude fetches the information from FreeScout and answers in the chat. Click the gray line above the answer to see which tools Claude used for it.

What Claude can't do with this option:

• Claude doesn't send any emails to customers. Drafts stay drafts until you send them yourself.
• Claude doesn't delete anything permanently and can't add new recipients to replies.
• Some actions, such as marking as spam, need approval by a person. Claude will then tell you that the action is waiting. You can approve it in the FreeScout menu AI approvals (KI-Freigaben) or directly in the conversation.

Claude Desktop. If you use the Claude Desktop app on Mac or Windows, you don't need to set anything up in addition. The app uses the same Claude account, and the “FreeScout” connector is automatically available there. Turn it on in the chat as described above.

Option 1 is done! ✅

Part C – Option 2: Claude as an agent in FreeScout

An administrator does this part. Allow about 30 minutes, followed by a few days of dry run.

What you need for Option 2

• Part A is done, and the self-check shows green for cron.
• Administrator rights in FreeScout.
• An account on Anthropic's Claude Platform (platform.claude.com) with credit – bought by credit card or, with a Max or Team plan, the monthly API credits included in the plan.
• An email address for the agent, for example james@deine-firma.ch. It doesn't need to receive mail anywhere, but it must not belong to a FreeScout user yet.

C1 – Creating an API key at Anthropic

With the API key, FreeScout uses Claude and pays for it from your credit at Anthropic. With a separate workspace and a spend limit, you make sure the agent never spends more than planned.

1. Open platform.claude.com and sign in. If you don't have an account yet, create one (company, name, email).

Credit

1. In the settings, under «Organisationseinstellungen» (organization settings), open Billing (Abrechnung).
2. Under «Guthaben» (credit balance) you can see how much money is in the account. If it's empty, click Buy credits (Credits kaufen) and buy, for example, 20 USD. With «Automatisches Aufladen verwalten» (manage auto-reload) you can have the credit topped up automatically when it runs low. That way the agent doesn't suddenly stop.

Workspace with a spend limit

1. Under «Organisationseinstellungen» (organization settings), open Workspaces (Arbeitsbereiche) and click Create workspace (Workspace erstellen). Name it, for example, FreeScout and confirm.
2. Click the new workspace in the list. An overview opens on the right.
3. Click Spend limits (Ausgabenlimits) there and set a monthly limit, for example 30 USD. Once it's reached, Anthropic stops the requests until the next month.

Creating the key

1. Under «Organisationseinstellungen» (organization settings), open API keys (API-Schlüssel) and click Create key (Schlüssel erstellen).
2. First, a window appears: «Mit Identitätsföderation brauchst du keinen API-Schlüssel» (with identity federation you don't need an API key). This is meant for large IT environments and doesn't fit FreeScout. At the bottom, click «Mit einem API-Schlüssel fortfahren» (continue with an API key).

1. Fill in the window «API-Schlüssel erstellen» (create API key):
   • Name: for example FreeScout Agent.
   • Expires (Läuft ab): choose Never (Nie). The default here is «30 Tage» (30 days) – then the agent stops working after a month without any warning. The yellow warning that appears with Never is correct: that's why you keep the key safe (step 11).
   • «Verknüpftes Konto» (linked account): leave your own name. Note the hint below it: if you leave the company and your access to the platform is removed, the key is deactivated too. A new key is then needed.
   • Scope (Geltungsbereich): choose the workspace from step 4, i.e. FreeScout. Then the spend limit applies to this key as well.
2. Click Create key (Schlüssel erstellen).

1. Copy the key right away. It starts with sk-ant- and is shown only this one time. Keep it safe until step C4, for example in your password manager – not in an email and not in a chat.

C2 – Creating a service account for the agent

The agent works as its own FreeScout user. Its name appears on every note and draft it writes.

1. In FreeScout, click Manage → Users (Verwalten → Benutzer) and then New User (Neuer Benutzer).
2. Fill in:
   • Role (Rolle): User (Benutzer), not Administrator
   • First Name (Vorname): the agent's name, for example James
   • Last Name (Nachname): for example KI, so everyone can see an AI is writing here
   • Email (E-Mail): the address for the agent, for example james@deine-firma.ch
   • Mailboxes (Postfächer): only the ones the agent should handle
3. Untick Send an invite email (Einladungs-E-Mail senden). The agent never signs in by itself.
4. Click Create User (Benutzer erstellen).

C3 – Creating the service client

The service client defines what the agent is allowed to do.

1. Under Manage → AI connections, click Agent on the left.
2. Scroll to Service clients (Dienst-Clients) and click New service client (Neuer Dienst-Client).
3. Fill in the window:
   • Name: for example James
   • Service account (Dienstkonto): the user from C2
   • Permissions (Berechtigungen): what the agent needs to start with is preselected: reading, notes, drafts, status and suggesting spam. Leave it like that for now. You can add more permissions later, such as “Answer questions in its own words” (Fragen in eigenen Worten beantworten) or the permissions of other modules.
   • Mailboxes (Postfächer): “All the service account can see” (Alle, die das Dienstkonto sieht) – you already made that choice in C2.
   • Allowed IP addresses (Erlaubte IP-Adressen): leave as is. It's prefilled with the addresses Claude works from at Anthropic.
   • Personal data (Personendaten): leave Pseudonymize (Pseudonymisieren) ticked.
4. Click Create service client (Dienst-Client anlegen).

The new service client appears in the list.

C4 – Setting up the agent and starting the dry run

At the top of the same page, in the Runner (Läufer) section:

1. Tick Process every new customer message (Jede neue Kundennachricht bearbeiten).
2. Tick Dry run (Probebetrieb). Important at the start: the agent then works on a trial basis but doesn't write anything in FreeScout.
3. Under Service client (Dienst-Client), choose the client from C3.
4. Under Anthropic API key (Anthropic-API-Schlüssel), paste the key from C1 (Mac: ⌘ + V, Windows: Ctrl + V).

C5 – Model, form of address and house rules

Right below:

• Model (Modell): by default claude-opus-5-5 with effort “medium”. That's thorough, but somewhat slower and more expensive. claude-sonnet-5-5 is faster and cheaper and is enough for many teams.
• Model in the live chat (Modell im Live-Chat): only if you use the live chat. In a chat, someone is waiting; a faster model or lower effort answers sooner.
• Form of address (Anrede): how the agent addresses the team in notes and how it addresses customers.
• House rules (Hausregeln): write in simple sentences who takes care of what, how you address customers and what is never promised. The agent bases its assignments and drafts on these.

C6 – Setting limits and saving

Under Limits (Grenzen) you set the most the agent will work and spend:

• Runs per hour (Durchläufe pro Stunde): protects against spikes, for example during a wave of spam.
• USD per day (USD pro Tag): the daily budget. Once it's used up, the agent pauses until the next day.
• Runs per ticket and day (Durchläufe pro Ticket und Tag) and Tool calls per run (Werkzeugaufrufe pro Durchlauf): stop the agent from getting stuck on one ticket.

Then click Save (Speichern) at the bottom. ✅

C7 – Watching the dry run

Let the agent run on a trial basis for a few days. For every new customer message it decides what it would do, but doesn't write anything.

1. Open Manage → AI connections → Log (Verwalten → KI-Verbindungen → Protokoll) and there the Agent runs (Durchläufe des Agenten) tab.
2. There you can see, for each message, what the agent would have done: the full text of the note and draft, the assignment and the cost.
3. If something isn't right, add to the house rules (C5).
4. Once you're happy, untick Dry run in C4 and save. From now on, the agent is working for real.

C8 – The emergency stop

At the very top of the Agent page is the Emergency stop (Notaus). Clicking Stop agent (Agent stoppen) halts the agent immediately, and all its access is blocked. With Release agent (Agent freigeben) it runs again.

C9 – Working with the agent

Once the agent is up and running, it works by itself. In addition, you can give it tasks:

• With a note in the ticket: write an internal note that starts with its name or mentions it, for example “@James, check the price again and rewrite the draft”. It takes care of it and replies with a note.
• In the AI menu: at the top of FreeScout there's the menu item AI (KI). There you give the agent tasks as in a chat, such as “Which customers have been waiting for more than five days?”. Tasks can also be repeated, for example as a weekly report.

• Approvals: if the agent wants to do something that needs approval, it appears under AI approvals and in the conversation. A person clicks Approve (Freigeben) or Reject (Ablehnen).

Option 2 is done! ✅

Part D – Checking or ending connections

D1 – Option 1: Viewing your connections in FreeScout

1. In FreeScout, click your name at the top right and then Your Profile (Ihr Profil).
2. On the left, click AI connections (KI-Verbindungen).

You see each connection with its creation date, last use, the allowed Permissions (Berechtigungen), the Mailboxes (Postfächer) and whether Pseudonymization (Pseudonymisierung) is on.

D2 – Option 1: Ending a connection

• In FreeScout: click Revoke (Widerrufen) next to the connection and confirm. Claude has no access from that moment on.
• In Claude: open Customize → Connectors, click the three dots next to the connector and then «Entfernen» (remove).

The safest way is to do both.

D3 – Option 1: Allowing more or less

The permissions of an existing connection can't be changed afterwards. Here's what to do:

1. Revoke the connection in FreeScout (D2).
2. Reconnect the connector in Claude, or remove it and add it again as in Part B.
3. Tick the boxes you want in the consent dialog.

D4 – Option 2: Pausing or ending the agent

• Stop immediately: Stop agent (C8).
• Just stop handling new customer messages: in C4, untick Process every new customer message. Tasks in the AI menu keep working.
• Change permissions: click Edit (Bearbeiten) next to the service client (C3).
• End completely: Disable (Sperren) the service client, remove the API key in C4 and also delete it on platform.claude.com under «Organisationseinstellungen» (organization settings) → API keys.

Part E – If something doesn't work

Option 1 – Claude Desktop and claude.ai

Option 2 – Agent in FreeScout

Security in brief

• Every connection has an owner. With Option 1, Claude works with your permissions; with Option 2, with the permissions of the service account – never with more.
• Write permissions are never preselected, except for the agent's basic setup, which you can see in C3.
• Sensitive actions need approval by a person; deleting permanently isn't possible at all.
• Personal data can be pseudonymized before it leaves FreeScout.
• Every new connection is reported by FreeScout by email, and every action is recorded in the log.
• One click is enough to end a connection or stop the agent.
• Only connect if you started it yourself. If someone asks you by email or phone to “quickly allow something”, click Cancel and report it to your administrator.