The marketplace
Your agents show up where your team already is
The short answer
It fits around how your team already works
Agents start themselves
Pick a time and it runs then. Or point it at an inbox, a folder, a form, a calendar, or a record changing in another system. It runs on the first of the month automatically, with nobody having to start it.
They answer where your team already is
Slack, Teams, Google Chat, Discord, Telegram, WhatsApp, SMS, email and Outlook. Plus Zendesk and Intercom when it is a customer asking, GitHub and Linear when it is a ticket, and a chat widget on your own site. Replies quote the Document behind them.
They plug into your systems
Connect your CRM, your ledger, your CMS — anything with an MCP server or an OpenAPI service. You choose exactly which operations an agent may call, and the credential you paste stays inside Steptree. It is never handed to the runner.
Whatever you already run
Four ways in, and the fourth is a screen
A connection you configure once
Point Steptree at an MCP server or an OpenAPI service and its operations become tools your agents can call. Or use the generic verbs against any HTTP API with a base URL and a key. That covers the long tail of business software that publishes no machine-readable description at all.
A tool built for the job
Where a wrong call is expensive, such as an advertising budget or a campaign going live, the integration is hand-built rather than generic. The dangerous operations are refused by name and the reversible ones sit behind an approval.
A channel your team already uses
Slack, Teams, Google Chat, Discord, Telegram, WhatsApp, SMS, email and Outlook. Plus Zendesk and Intercom when it is a customer asking, GitHub and Linear when it is a ticket, and a chat widget on your own site.
A screen, when there is no other way in
The in-house system, the desktop-only application, the twenty-year-old program the whole business depends on. The desktop runtime works it with a keyboard and a mouse, which is the only integration those will ever have.
You find out in the first five minutes, not after a fortnight
Your keys, your rules
Connecting something once does not hand it to everybody
- The credential belongs to the workspace
- Connect a system once for the business, not once per agent. A rotated key is encrypted and lands in one place, and every agent using it picks it up — instead of three stale copies nobody can find.
- It never leaves Steptree
- Connections are resolved here and the tools are handed to the runner already namespaced. Your key is encrypted, the cloud runner holds no tenant credential, and it cannot end up in somebody's transcript.
- Teams decide who may use it
- A connection can be granted to the whole workspace or to named teams. The grant is re-checked when it is used, not when it was attached, so narrowing access takes effect immediately.
- Each agent is attached deliberately
- Having a connection does not give every agent access to it. A manager ticks the box per agent. Where several connections could serve one tool, the agent says which — you choose which one it spends from, not whichever comes first alphabetically.
- An unusable credential says so
- Switched off, revoked or incomplete, it is reported before a run rather than silently falling back to some older copy. That failure mode cost an operator a day of a rotated token appearing not to work.
- You choose the operations
- A connected system does not mean a blank cheque. You pick which operations an agent may call, and any of them can be set to stop and ask a person first.
Questions
About connecting things
- Do you integrate with [our system]?
- Probably. Here is how: Steptree recognises around 170 applications. For each one it records how a cloud agent reaches it: a channel, a built-in tool, a connection, a plain HTTP call, or only through a desktop screen. Anything with an API you can call is reachable today without us writing vendor-specific code.
- How do I know before I commit?
- Steptree reads your recording, recognises the applications in it, and tells you on the recording itself which of them a cloud agent can reach and which need a desktop.
- Where do our API keys live?
- In your workspace, encrypted, resolved by Steptree at the moment a tool is called. The cloud runner never holds them, and credentials are redacted out of what a model sees. You can rotate one in place without detaching every agent using it.
- Can an agent start when something changes in another system?
- Yes — Steptree polls the systems you have connected and starts the agent when something it cares about moves. That turns any connected application into a trigger with no per-vendor code on your side.
- What about a system with no API whatsoever?
- The desktop runtime. An agent drives it with a real keyboard, mouse and screen on your own machine, with an approval card before each action. That is the half of most businesses with no API at all.
- Can we use our own MCP server?
- Yes. Point Steptree at it, and its published tools become tools your agents can be granted. Steptree also works the other way round: your Documentation can be published to the MCP client you already work in, on every plan.
Next
Where to go from here
What agents do with it
The capabilities on the other end of those connections, and the limits you hold them under.
Three worked examples
A finance close, a support queue and a system with no API, told end to end.
How credentials are handled
Where keys live, what the runner can see, and what is never sent to a model.
Do one real job once, and read the SOP Steptree writes
Record it on Windows or Mac. Steptree writes the SOP with the goal, the steps and the judgement calls. Approve it, share it with your team, and connect your AI. Free to start, no card.
