Quickstart

Install superagnt for me. It's the superagnt plugin from the github.com/superagnt/plugins marketplace, plus the superagnt MCP server (on Claude it's a connector I add in my browser; elsewhere I sign in through my browser). The setup guide for each kind of agent is at https://www.superagnt.com/install

The agent fetches the served markdown at that URL, picks the branch that matches the client it is running in, writes the config, and runs the browser consent flow. Because the instructions are a live URL and not a snapshot, a client fix on our side repairs every copy of the line that was ever pasted.

Once it is connected, the doc tells the agent to make agnt_onboarding its first tool call. That is the tool that records what you want automated and hands back the platform brief, so the agent can tell you the handful of things worth doing first instead of reciting a catalog.

Before you start

You need a workspace, and then either an OAuth-capable MCP client or a workspace API key. Sign up at app.superagnt.com. Onboarding mints a Default key for the new workspace; you reveal that one, or create another, from the dashboard.

Connect a client by hand

On Claude the MCP server is a connector: open the prefilled dialog on claude.ai and click Add. Claude Code signed in with your Claude account picks it up in its next session. Then install the plugin for the superagnt skills, the know-how to use the tools.

/plugin marketplace add superagnt/plugins
/plugin install superagnt@superagnt

On Claude Code signed in with an API key, connectors never load, so add the server directly with the manual pair (no skills). --scope user is not optional: the CLI defaults to project scope, which hides the server from every directory except the one you ran it in.

claude mcp add --scope user --transport http superagnt https://mcp.superagnt.com/mcp
claude mcp login superagnt

Every other client has its own one-liner or config block. Connect your client has all of them, including the three whose config keys differ enough to fail silently when copied from another client.

Call the REST API

Start with the catalog. GET /v1/platforms needs no credential at all and is the authoritative, always-current list of sources, endpoint counts and per-call prices.

curl -sS https://api.superagnt.com/v1/platforms

Then make one authenticated call. The public API accepts Authorization: Bearer <key> and nothing else.

export AGNTDATA_API_KEY="agnt_live_..."

curl -sS "https://api.superagnt.com/v1/data/linkedin/get-company-by-domain?domain=apple.com" \
  -H "Authorization: Bearer $AGNTDATA_API_KEY"

Deploy your first agent

The third path does not run on your machine at all. You describe the job to the Agnt Wizard in the dashboard, it builds an Agent, and the Agent runs on our infrastructure against the same workspace.

1. 01

   Open the builder

   Go to app.superagnt.com and tell the Agnt Wizard what you want done. It drafts the Agent, its tools and its system prompt with you.
2. 02

   Give the draft a real name and a system prompt

   Deploy refuses a draft still called "Untitled agent" and refuses an empty system prompt. Both come back as a 400 VALIDATION_ERROR naming which one is missing.
3. 03

   Clear the deploy gate

   Deploying needs an active plan, a managed plan trial, the no-card trial (until it ends) or an active credit package. Without one, deploy returns 402 DEPLOY_GATE_FAILED and the dashboard offers a plan instead of a dead end. On a builder plan or the trial, deploying past the plan's agent limit returns an offer for the next plan.
4. 04

   Watch it run

   A deployed Agent gets its own key, its own sessions and its own observability trail. Everything it spends lands in the same two organization wallets you were already reading in meta.