Resources · Claude connector

Claude Sprites connectorWhat Claude can do in your Sprites account.

The Claude Sprites connector exposes 18 tools. 8 read the state of your Sprites, 10 change something, and none is left undescribed. Below: what Claude starts, what it can wipe for good, and the restricted token that fences it all in.

Verified Trustpilot reviews · AI, automation & growth agency

Overview

What changes for your coding experiments

A Sprite is an isolated, persistent Linux environment that Fly.io hosts for you. With this connector, you ask Claude to spin up a machine, run a command in it or save its state, and it does so from the conversation. Fly.io presents Sprites as the simple answer to the question of where to run a piece of code.

Test in an isolated environment. create_sprite opens a fresh environment, then exec runs the command you ask for, such as checking which runtime version is installed. Fly.io gives this kind of request among its own examples.

Undo a bad move. checkpoint_create freezes the filesystem before a change; if an install goes wrong, checkpoint_restore rewinds the Sprite to that point.

Run a service and keep an eye on it. Claude creates, starts or stops a background service, then reads its recent logs through service_logs.

Worth knowing first: destroy_sprite deletes a Sprite and its data with no way back, and a restore wipes out everything done after the chosen point. By default, the token only allows Sprites named with the mcp- prefix, and only a limited number of them. Nothing runs by itself either: Claude fires no command on a schedule. That's an automation tool's job, a different trade; the Integrations hub covers that side.

Vocabulary

The vocabulary in one minute

Five words you'll meet while connecting Sprites.

Connector
The link you set up once between Claude and an account you already have, so Claude can work in it while it answers you.
Tool
One named action a connector opens to Claude. Claude picks the ones it needs on its own, and the directory sheet lists them by name.
Authorization
The service's own sign-in screen, where you hand Claude the access it will use. Granted once per person, and you can take it back later.
Approval
The confirmation Claude waits for before it goes through with something that changes your account, shown in the chat when it matters.
MCP
The shared standard connectors are built on: it is what lets an assistant like Claude talk to an outside service in the first place.
Connect

Connect Sprites to Claude in three steps

  1. 01

    Find Sprites in Claude

    In Claude's settings, open Customize, then Connectors, and look for Sprites in the list. On a Team or Enterprise workspace, an Owner or Primary Owner enables the connector first, and each member can then connect with their own account.

  2. 02

    Start the connection

    Click Connect on its row, sign in with your Fly.io account in the window the service opens itself, and pick the organisation to use. If the link breaks, use Disconnect and connect it again.

  3. 03

    Read the authorization screen

    Read the authorization screen before accepting: it belongs to Fly.io, not Claude, and it sets what the access covers, token restrictions included. Whatever you grant there can also be withdrawn from your Fly.io account.

Tools

The 18 tools, grouped by what they do

Sprites gives Claude 18 tools: 8 that read your account, 10 that change something in it.

Two groups: what Claude reads, and what it changes in your Fly.io organisation. Names stay exactly as Claude shows them.

  • 8 read
  • 10 write

What Claude reads (8)

8 tools

Eight tools that look at your Sprites, their checkpoints, services and network rules.

checkpoint_get

Shows the details of one specific checkpoint on a Sprite, so you know which state it captures before deciding anything. Fly.io names inspecting as one of the checkpoint gestures, separate from listing them.

When it helps
you want to know which filesystem state a checkpoint holds before restoring it.

Sourcedocs.sprites.dev · October 1, 2026 ↗

checkpoint_list

Lists the checkpoints that exist on a Sprite, with their IDs, so Claude can then inspect or restore one of them later on.

When it helps
before any restore, you want to see every frozen state available on your test environment.

Sourcedocs.sprites.dev · October 1, 2026 ↗

exec_list

Shows the exec sessions open in a Sprite without interrupting any of them. Fly.io groups it under exec, next to running a command and killing a session.

When it helps
a long-running command looks stuck and you want to know whether it's still going before you kill it.

Sourcedocs.sprites.dev · October 1, 2026 ↗

list_sprites

Lists the Sprites visible to the connected organisation, within the token's policy. It's the right first step whenever a name doesn't respond.

When it helps
you want to know which of your mcp- Sprites are running right now.
Watch out
a Sprite outside the allowed prefix or in another organisation won't show up.

Sourcedocs.sprites.dev · October 1, 2026 ↗

policy_network_get

Reads a Sprite's outbound network rules, so you know where it's allowed to connect and what is blocked.

When it helps
a download fails inside a Sprite and you want to check whether the outbound rule forbids it before looking elsewhere.

Sourcedocs.sprites.dev · October 1, 2026 ↗

service_get

Inspects one background service inside a Sprite to see where it stands, without starting or stopping it. Fly.io names inspecting a service as its own gesture, apart from listing services.

When it helps
you want the state of one background service in a Sprite.

Sourcedocs.sprites.dev · October 1, 2026 ↗

service_list

Lists the background services defined in a Sprite, the starting point before inspecting, starting or stopping one. Fly.io generates these Sprite tools from the environment’s public API, so the set can evolve with new versions.

When it helps
you want to see which services a Sprite has before acting on one of them.

Sourcedocs.sprites.dev · October 1, 2026 ↗

service_logs

Pulls the recent logs of a background service, so you can see what the service reported. Fly.io describes this gesture as reading a service’s recent logs, nothing older.

When it helps
a service in a Sprite has failed and you want its recent log lines.

Sourcedocs.sprites.dev · October 1, 2026 ↗

What Claude changes (10)

10 tools

Ten tools that act on your Sprites. No source gives a per-tool confirmation, so the general rule below applies.

checkpoint_create

Approval: see the rule

Freezes a Sprite's current filesystem into a checkpoint you can return to later.

What Claude asks for
no source describes a confirmation specific to this tool; the general approval rule is what applies.
When it helps
right before installing a package you're unsure about, just as Fly.io suggests in its own examples.

Sourcedocs.sprites.dev · October 1, 2026 ↗

checkpoint_restore

Approval: see the rule

Rewinds a Sprite to the filesystem state of an earlier checkpoint. Everything done since that point is gone.

What Claude asks for
there's no documented rule for this exact tool, so the default approval rule applies.
When it helps
a change went wrong after a checkpoint and you want the Sprite back to that saved state.

Sourcedocs.sprites.dev · October 1, 2026 ↗

create_sprite

Approval: see the rule

Opens a new Sprite in the organisation chosen at connection time, ready to take commands and services.

What Claude asks for
Fly.io describes no specific confirmation here; the general approval rule applies.
When it helps
you want a separate, isolated environment to try something in.
Watch out
with the default token, the name must start with mcp- and the count is capped.

Sourcedocs.sprites.dev · October 1, 2026 ↗

destroy_sprite

Approval: see the rule

Permanently deletes a Sprite and all its data. Fly.io is blunt about it: there is no undo.

What Claude asks for
no rule of its own is documented; the general approval rule governs this tool.
When it helps
you're clearing out test environments nobody uses anymore.
Watch out
confirm the exact name with list_sprites before you agree.

Sourcedocs.sprites.dev · October 1, 2026 ↗

exec

Approval: see the rule

Runs a command inside a Sprite. Per Fly.io, a command can modify files, install packages, start services or use whatever network access the Sprite's policy allows.

What Claude asks for
no source describes a dedicated confirmation; the general rule applies.
When it helps
you want to install a command-line tool and then check its version.

Sourcedocs.sprites.dev · October 1, 2026 ↗

exec_kill

Approval: see the rule

Ends an exec session running inside a Sprite.

What Claude asks for
nothing tool-specific is described by any source; it falls back on the general approval rule.
When it helps
an exec session running in a Sprite should not keep going.

Sourcedocs.sprites.dev · October 1, 2026 ↗

policy_network_update

Approval: see the rule

Updates a Sprite's outbound network rules, to allow or block access to the outside.

What Claude asks for
no confirmation is described for this tool; the default approval rule holds.
When it helps
an install needs to reach a package registry the current rule blocks.
Watch out
opening the network also widens what any later command can reach.

Sourcedocs.sprites.dev · October 1, 2026 ↗

service_create

Approval: see the rule

Declares a new background service inside a Sprite, which can then be started, stopped or inspected.

What Claude asks for
no specific rule is published for this tool; the general approval rule covers it.
When it helps
a process should keep running in the background inside the Sprite.

Sourcedocs.sprites.dev · October 1, 2026 ↗

service_start

Approval: see the rule

Launches a background service already declared in a Sprite so it begins running.

What Claude asks for
the source says nothing about a particular confirmation; the general approval rule applies.
When it helps
a service already defined in the Sprite is stopped and you want it running.
Watch out
a cold Sprite can take a moment to wake.

Sourcedocs.sprites.dev · October 1, 2026 ↗

service_stop

Approval: see the rule

Stops a background service running in a Sprite without removing its definition.

What Claude asks for
no rule of its own is documented; the default approval rule is the one in force.
When it helps
a test service is running for nothing and you'd rather switch it off until the next session.

Sourcedocs.sprites.dev · October 1, 2026 ↗

Approvals

What Claude asks you before acting

By default, Claude stops and asks for your go-ahead before every action it takes on an account for you. The request shows up in the conversation, right when it matters.

Fly.io annotates its tools as read-only or destructive, and notes that the client may ask for extra confirmation before a destructive tool, depending on its settings. On a Team or Enterprise workspace, owners decide whether a member may let some actions through without being asked again, and they can cap the connector for the whole organisation. Finally, Claude acts with the token you approved and nothing more.

Plans

Which plans it's available on

Of the 819 sheets in the official directory, none shows availability by plan. For Sprites as for the rest, the connector-by-connector answer isn't published anywhere.

The general rule, though, is published: remote connectors are open to all users on Claude, Cowork, Claude Desktop and mobile. On Team and Enterprise, an Owner or Primary Owner turns the connector on for the organisation before each member can connect. For the current state, check the connector's sheet in the official directory.

Limits

Where this connector stops

A connector isn't an automation: Claude calls these tools while it answers you, and nothing starts on its own inside a Sprite. Only the publisher, Fly.io, documents this connector.

The tool list is an observed floor, not a promise, and Fly.io warns that the Sprite-level tools can change as new versions of the environment ship. The partner badge isn't a security audit, and Anthropic says so on every sheet: it doesn't choose a publisher's tools or vouch for how they behave. Connect only what comes from a publisher you know, here Fly.io. Other connectors are listed on the Claude integrations page.

Need help

Need help connecting Sprites to Claude?

A person reads every message.

FAQ

FAQ about the Claude Sprites connector

01What can Claude do with the Sprites connector?
Claude can create, drive and clean up Sprites, Fly.io's isolated and persistent Linux environments, straight from the conversation. Its eighteen tools cover creating and destroying Sprites, running commands, checkpoints, the outbound network policy and background services with their logs. Eight tools only read the state of your Sprites, the other ten act on them. Fly.io describes a Sprite as a persistent Linux computer, isolated at the hardware level, for running arbitrary code.
02Can Claude delete or change my Sprites?
Yes, ten tools act on your Fly.io organisation. Claude can create and destroy a Sprite, run or kill a command in it, create and restore a checkpoint, update the network policy, and create, start or stop a service. Destroying wipes the Sprite and its data with no way back, and a restore erases the work done after the chosen point. The default restricted token limits creation to the mcp- prefix and a capped number.
03Does Claude ask before destroying a Sprite?
By default, Claude asks for confirmation before every action it takes on an account for you, and no source describes a rule specific to destroy_sprite or the other tools. Fly.io adds that its tools are annotated as read-only or destructive, and that the client may ask for extra confirmation before a destructive tool depending on its settings. On Team or Enterprise, owners decide whether members can let some actions through without a prompt.
04Which plans is it available on?
No official source publishes plan availability connector by connector, and no directory sheet shows it. The general rule is that remote connectors are open to all users on Claude, Cowork, Claude Desktop and mobile, with an Owner or Primary Owner enabling them first on Team and Enterprise. The connector's sheet in the official directory remains the only place that shows its current state for your own account, so check it before relying on the connector.
05Does Claude see our whole Fly.io organisation?
No, it sees what the token you approved at connection opens up. In restricted mode, the default one, the connector only creates Sprites with the chosen prefix and within the cap, and list_sprites only shows what the token policy allows. Full access, on the other hand, lets it manage every Sprite the organisation token can reach. Fly.io recommends sticking to restricted mode unless you genuinely need broader control over the organisation.
06What if Claude can't find a Sprite?
According to Fly.io, a missing Sprite may belong to a different organisation, sit outside the prefix the token allows, or have been destroyed already. Ask Claude to call list_sprites to check the exact name. If you belong to several organisations, make sure you picked the right one when connecting. And if the Sprite was cold, it can take a moment to wake: a few seconds of waiting is enough before running the same tool again.
07Claude or an automation tool for Sprites?
They answer different needs, so it comes down to the task. Claude works inside a conversation: you ask for an environment, a command or a checkpoint, Claude calls the tools while it answers, and nothing else happens afterwards. It reruns nothing on a schedule and doesn't react to events on its own. For a scheduled job that runs in the background, an automation tool is the right fit. A request made in the moment belongs to the conversation.