Tools

Tools lists what your agents can reach for: the packages the platform runs for everybody, and the MCP servers you connect yourself.

The Tools catalog with Google Workspace, Linear, Notion and others — Browse: the MCP catalog, with the transport and the kind of authentication on each card

Three tabs. Browse is the catalog of MCP servers you can connect, System tools is what you already have, and Your MCP servers is what you have connected so far.

System tools

Six packages, on for every agent on the account unless you turn one off:

Package	What it is for
Media	Read images, audio, video and documents — binary in, text or structured data out.
Search	Search the live web and pull page contents back as text.
Email	An address of your own: read the mail sent to it, send mail from it.
SMS	A real phone number to receive SMS on, for a signup that asks for one.
Memory	Search everything said in past chats, and the window the run is carrying now.
Harness	The questions about a harness that its own checkout cannot answer.

Each card says Available in every run and offers Turn off.

What `system:` means

The slug on that screen is the second half of a name you write in code. An agent gets a tool by naming its source in its declaration:

assistant = agent(
    "Assistant",
    system_prompt=prompt("assistant"),
    tools=[
        "system:compute",                          # a platform package
        "system:search#search",                    # one tool out of a package
        "tools/files.py",                          # a file of your own harness
        "019e4f9a-a679-79b4-a21c-7ff9c60b9181",    # an MCP server you connected
    ],
)

system:<slug> is a package the platform runs, credentials and all. You cannot edit one and you do not publish anything for it to work. A # after the source narrows it to named tools and nothing else.

system:compute — the runner every other tool executes through — is not on this screen, and neither is system:splox. They are sources you can name in code without being listed here as something to switch. The full list of sources your account can use comes back from /v2/tool-servers.

So there are two switches, and they do different things. This screen decides what your account has; the tools list in the harness decides what a given agent gets. An agent is only ever offered what its own line names. Tools is the whole story, including writing your own.

Connecting an MCP server

Pick a server from the catalog and Splox shows what connecting it takes. The badges on the card said it already: oauth or manual for the credential, http or sse for the transport.

The connect page for the Linear MCP server — An OAuth server: you authorize at the provider, and Splox never sees the password

OAuth — name the connection and press Connect. You are sent to the provider to authorize, and back.
Manual — the page asks for exactly the headers that server wants, named one by one (X-api-key, X-access-token, and so on), with a link to where you get them.

A server that is not in the catalog is Add MCP server:

The custom MCP server form with URL, transport and headers — Adding a server the catalog does not have

Give it a name and a URL, pick the transport, and add whatever headers it authenticates with. It then appears under Your MCP servers like any other, and an agent gets it by putting its id in tools.

Tools in the harness

Writing a tool of your own, and why tools are code and not JSON calls.

Tool servers over the API

Listing the sources and the tools inside them.

Start here

Ask your agent

Look inside

Concepts

In the app

Account

Help

System tools

What `system:` means

Connecting an MCP server

Tools in the harness

Tool servers over the API

​System tools

​What system: means

​Connecting an MCP server

Tools in the harness

Tool servers over the API

System tools

What `system:` means

Connecting an MCP server