# tapinall > One token for thousands of agent tools. An agent searches the catalog by > task (catalog_search), inspects a tool's method and price (catalog_get), > and runs it (call) — per-call pricing against a prepaid credit balance, > shared between the CLI and MCP. ## Set up tapinall for the human's agent host 1. Install the CLI (macOS / Linux): curl -fsSL https://tapinall.com/install.sh | sh Follow the installer's PATH instruction. On Windows, download tapinall.exe and tapinall-mcp.exe plus NOTICE and LICENSE from https://tapinall.com/v1/cli/download/windows-x64?bin=tapinall (bin=tapinall-mcp for the forwarder) into one folder on the user's PATH. 2. Sign in — this is the human's step, not yours. The catalog needs a prepaid account. Ask the human to sign in at https://tapinall.com/v1/auth/login, add credits, and generate a project token (tap_…). Then store it: tapinall login # prompts for the token tapinall login --token tap_… # or: echo "$TOKEN" | tapinall login 3. Register the tools in the host you run inside: tapinall mcp install --host claude # also: cursor, codex, opencode, gemini — or --host all Other agents (Hermes, OpenClaw, Pi, any MCP client): register a stdio server with command `tapinall-mcp` and no arguments, after login. If the agent runs from a desktop app or service, use the binary's absolute path (`command -v tapinall-mcp`) — that environment's PATH may differ from the terminal's. Remote-MCP-only platforms: POST https://tapinall.com/mcp — the same three tools over OAuth; tapinall is its own authorization server (clients register and run code+PKCE against https://tapinall.com/.well-known/oauth-authorization-server). ## After setup The host exposes exactly three tools — catalog_search, catalog_get, call — the same verbs as the CLI (tapinall catalog search|get, tapinall call, tapinall balance, tapinall usage ). Prices are quoted per call before execution; rejected requests and provider errors are not charged.