Agent Tools
Reference for the surfaces an AI agent uses to build, run, and pay for KeeperHub workflows. To connect an agent for the first time, start with Getting Started with an Agent.
| Surface | What it does | Best for |
|---|---|---|
| MCP Server | Model Context Protocol server with more than forty tools covering workflow CRUD, execution, and direct onchain writes | AI agents, custom integrations, remote automation |
| Claude Code Plugin | Skills and slash commands on top of the MCP tools | Developers working inside Claude Code |
| Agentic Wallets | An x402 / MPP wallet so an agent can pay for workflows | Agents calling paid marketplace workflows |
Authentication
The MCP server accepts either credential on each request:
- An OAuth access token, minted in the browser (for example
/mcpin Claude Code). Short-lived and the right choice for an interactive agent. - An organization API key (
kh_), created under your avatar, then API Keys, then the Organisation tab. Long-lived and the right choice for headless, CI, and Docker.
User-scoped keys (wfb_) authenticate webhook triggers and do not work here. See
API Keys.
Authoring and testing workflows
- MCP Trigger Inputs - the input schema each trigger type expects
- Validate Workflow - catch structural problems before saving
- Test Workflow - pin test data and dry-run a workflow
Running and reading executions
- Get Execution - the
{ status, logs }response shape, which fields are numbers versus strings, and what is redacted cross-organization