Connect Island Browser

Island is an enterprise browser. It is a Chromium browser with policy, data-loss prevention, session control, privileged access and browsing telemetry built into the browser itself, all managed from the Island Management Console. StackJack talks to that console through its own API, which is the same interface the console uses.

Connecting Island Browser to StackJack gives your AI assistant a family of island_ MCP tools — MCP (Model Context Protocol) tools are the standardized commands an AI assistant can call through StackJack. With them, your AI can:

• Answer "who and what" questions — browser users and their status, devices and the extensions installed on them, the extension risk profile for any extension in the estate, and the roles that govern your console
• Report on experience — Island's Digital Experience scores for applications and devices, with speed, stability, usage and the errors behind a poor score, broken down however you ask
• Surface security findings — compromised credentials Island has detected, the insights it raises and the issues behind them
• Read the audit trails — the browser event timeline and the record of what administrators changed in the console
• Review policy before it ships — what is staged but not yet pushed, whether a rollback is still possible, and the history of published policy
• Manage the plumbing — connector appliances and their groups, the commands sent to them, notification policies and who they deliver to, and SIEM forwarding
• Administer tenants and people (on paid plans) — create and update tenants, invite users, modify or remove them, block and unblock a person or a machine, move users in and out of roles, and push or discard staged policy
• Read and edit the policy surface itself (on paid plans for the edits) — the browser experience and security settings profiles, the application access, cloud firewall, SSL inspection and AI Guard rule sets, DLP profiles, detectors and data locations, and the URL, network, identity, location and device-posture objects the rules are built from
• Run the workspace your people see (on paid plans for the edits) — branding, the homepage and its sections and shortcuts, the application catalogue and application groups, custom end-user messages, and privileged application access
• Take a portable copy of anything — every configuration object has an export, which comes back either as the object itself or as a temporary download link when Island answers with a file. The twenty-six exports that carry credentials are handed back as the object only, never stored as a file

How StackJack authenticates to Island

Island gives you one thing: an API key. You paste it into StackJack and pick your region. There is no client ID, no second secret, and nothing to renew on a schedule.

An Island administrator creates the key in the Management Console under Modules > Platform Settings > System Settings > Integrations > API. Choose the API Key type rather than the OAuth type.

The Role you pick at creation is the real limit

When you create the key, Island asks you to choose a Role. That Role is what the key can do, and nothing StackJack does can widen it. Give the key a read-only Role and StackJack can report on everything and change nothing, no matter which tools you enable. Give it a full-admin Role and the write tools become genuinely capable.

This is the most useful control you have, and it is worth spending a minute on. Most people want a read-only key first, and a second key with more authority later if they decide they want an assistant to act.

Pick your region

Island runs three separate regional services: United States, Europe and United Kingdom. They are independent — a key created in one does not work in another.

StackJack asks which one your tenant is on and stores that choice, rather than storing a web address, so the connection cannot drift onto the wrong service later. Pick the region whose Management Console you signed in to when you created the key.

If you pick the wrong one, Island refuses the key with an error that looks exactly like a bad key. If a brand-new key is refused, check the region before you create another one.

Steps

1. Sign in to the Island Management Console with an Admin or System Admin account.
2. Go to Modules > Platform Settings > System Settings > Integrations > API.
3. Create a new key of the API Key type and choose the Role it should carry.
4. Copy the key. Island shows the value once, at creation, and never again.
5. In StackJack, open the Island Browser connector, select your region, paste the key, and save.

One key, one Island tenant

This is the most important thing to know if you manage several customers in Island.

Island's API does not take a customer selector on any call. A device list, a user list, a policy read and a Digital Experience report all answer for the tenant the key belongs to, and there is no way to ask one key about a different tenant's data.

So an MSP adds one connection per customer: one Island key, one region, one named connection. Your AI names the connection on each call; omit the name and the call runs against your default connection. Pin an endpoint to one connection when an assistant should never reach past a single customer. See Several connections of one connector.

Island does have a tenant hierarchy, and the connector can read and administer it — list the tenants a key can see, create a sub-tenant, set who it admits. What it cannot do is read a child tenant's devices, users or policy with the parent's key.

What to know before you turn it on

Reads change nothing, even though many of them look like writes

A large part of this API works by sending a request body rather than by reading a web address. Every Digital Experience report, the device list, the compromised-credential list and both audit searches send data to Island in order to describe the filter they want. They still create nothing and change nothing.

StackJack classifies these by what they actually do rather than by how they travel, so they are Free-tier reads and are not marked destructive. They do return your users, their devices and their browsing telemetry, so grant them deliberately.

Some tools act on a live browser fleet

Five of them deserve a closer look before you let an assistant near them:

• Pushing policy applies every staged change at once, to every managed browser. There is no partial push. Ask your assistant to read the pending changes first and confirm the list with a person before the push.
• Discarding staged policy throws the work away. There is no undo, and the check for whether a rollback is still possible is a separate read worth running first.
• Blocking a user or a machine takes the browser away from someone who is working, immediately. Forcing a logout ends live sessions and loses unsaved work in them.
• Registering a connector hands back a registration token. That token is a credential; treat the reply as one.
• Testing a notification webhook sends a real notification. If the subscriber group pages a human, a test pages them.

Every one of these is marked destructive. Whether your AI application asks you to confirm before running one depends on that application's own settings — see Destructive tools and confirmation. Review that setting before you enable the write tools.

Two tools read as destructive from the name and are deliberately not: unblocking a user and unblocking a machine both restore access rather than removing it.

Modifying a user can also delete one

Island's modify-user call accepts flags that delete or revoke the user in the same request as an ordinary edit. StackJack marks it destructive for that reason. If you only want to change a detail, be specific with your assistant about which fields to send.

Updating a SIEM forwarder replaces the whole configuration

There is no partial update. A field left out will most likely be cleared, and the record includes the markers Island uses to track how far it has forwarded — so an incomplete update can quietly rewind or skip log delivery. Read the current configuration first and send it back complete.

The whole Management Console is covered, and that is a lot of tools

StackJack now reaches every part of Island's Management Console API: tenants and roles, users, devices and extensions, Digital Experience, security findings and audit, policy staging and rollout, connectors, notifications and SIEM — and the policy surface underneath all of it. The browser experience and security settings profiles, the firewall, SSL-inspection and AI Guard rule sets, DLP profiles and detectors, URL, network, identity and device-posture objects, the branding, homepage and application workspace, API keys and external integrations.

That is more tools than some AI assistants will accept at once — Claude Desktop, for instance, stops at 256. If your assistant starts losing tools or refusing to connect, turn on catalog modes for your organization on the Settings page and pin the tools you actually use. That gives the assistant a short list it can hold while leaving everything else reachable on request; it is a better answer than turning parts of the connector off.

Exports hand back a download link

Island's own API description does not say whether an export answers with JSON or with a file, and StackJack does not guess: it passes Island's answer straight through when it is JSON, and when it is a file it stores it and hands your assistant a temporary download link instead. The link is short-lived by design — about thirty minutes — so fetch it when you get it rather than saving it for later.

Twenty-six exports are the exception, and they carry credentials. The API-key exports, the privileged-application-access exports, and the AI Guard, DLP 360 and connector-group exports answer with live secrets. StackJack passes those through unchanged when Island answers with JSON, and refuses them when Island answers with a file: it does not store a file of live credentials or hand back a link to one. Export those from the Island console when you need the file.

A sign-in page is not an export. If a proxy or a captive portal answers in place of Island — a web page where the export should be — StackJack reports the failure instead of storing the page and handing you a link to it. Your Island credential is not at fault when that happens; check what sits between StackJack and Island.

Large reads

Where a read accepts a size, Island publishes no maximum and StackJack adds none, so ask for what you need. Most large reads take a filter and page with a size and an offset; there is no cursor to follow. Keep audit windows tight — a wide one returns everything in it.

Plans and limits

Read tools are available on the Free tier — every one of them, the exports included. Taking a portable copy of a policy, a rule set, a profile or the whole tenant configuration costs nothing, whether Island answers with the object itself or with a download link (and for the twenty-six credential-bearing exports, with the object itself only). Every change needs Pro — creating a tenant, updating a connector group, creating a notification policy, unblocking someone, and equally the tools that act irreversibly or on a live fleet: pushing and discarding policy, blocking, deleting, inviting, granting a role, registering a connector, running connector commands, and replacing a SIEM forwarder. That second group is marked destructive, so an assistant that honors the label asks before running one. Business offers the same tool set as Pro with a higher monthly call quota.

See the generated Island Browser tool reference for the current inventory, plan assignment, input schemas, and destructive-action labels.

Island publishes no rate limit, so StackJack paces requests conservatively and backs off on its own if it is throttled, which usually makes a large report slower rather than failed. Pacing smooths a burst; it does not guarantee that every call arrives. Retries are bounded, so a wide enough read can still come back throttled or time out. Narrow the read, honour any retry delay the vendor sends, and check whether a write landed before repeating it — see Retrying a failed or timed-out write.

Troubleshooting

"Island Browser rejected the API key" — three causes, in the order worth checking. The region is wrong; the key was deleted or regenerated in the console; or the key was pasted incompletely. A key created against the United States console and used with Europe selected fails exactly this way, and it is the easiest of the three to miss because the message is the same.

"Island Browser denied the request" — this is the Role, not the key. The key is valid and Island understood the request; the Role chosen when the key was created does not allow that operation. Either use a tool that only reads, or have an administrator create a key with a wider Role. Ask your assistant to list your roles if you are not sure which one the key carries.

"Island Browser could not find that object" — Island's ids belong to one tenant, and one key reaches one tenant. An id copied from another customer's connection will not resolve. List first, then act on an id from the same connection.

A connection test passes but every device read comes back empty — the key is valid and the Role allows the read; the tenant genuinely has no matching devices, or the filter is narrower than you meant. Try the same read with no filter before assuming a permission problem.

A policy push reports success but nothing changed on a browser — the push is asynchronous. It returns an identifier, and there is a separate read that reports the status of that push. Ask your assistant to check it rather than pushing again; a second push applies the same staged changes over the top.

Island Browser tools

island_ · 1012 tools · Free 595 · Pro 417

Tenants and Roles

Users

Devices and Extensions

Digital Experience

Security and Audit

Policies

Connectors

Connector Groups

Notifications and SIEM

Browser Experience Settings

Browser Security Settings

Policy Rules

Data Protection

Policy Objects

Workspace and Branding

Administration and Integrations