Cubeshell Docs

Cubeshell is a single MCP connection your AI agent uses to call paid tools on demand. One connection, one prepaid credit balance, no per-tool subscriptions.

How it works

Create a workspace in the console, create its MCP connection, and add that connection's URL to Claude, Codex, or another MCP-compatible client. Every call your agent makes goes through that one connection, is priced before it runs, and is charged in microcredits against your workspace balance.

Tools never take a workspace ID as an argument — the connection URL itself identifies the workspace, so an agent can never accidentally call a tool against the wrong one.

Connection URL
https://api.cubeshell.com/mcp/c/cn_your_connection_id

Tools available today

Cubeshell exposes 14 tools across 5 families. Each one has its own reference page with the exact input it takes and the result it returns.

Your MCP client's tool list is always the authoritative set: every listed tool is callable, and each carries its own input schema and price. The web tools work on one exact public URL; Cubeshell does not expose a web-search tool.

i
Execution disclosure
Every result identifies its execution mode. Email verification may use the launch verification route, while web scrape and screenshot use live bounded execution. Price, workspace, operation, replay status and receipt remain explicit on every result.

Credits

One denomination (credits-v1) is used everywhere: 1,000,000 microcredits per displayed credit. The launch starter pack is $49 for 1,000 displayed credits. Every tool call returns the exact price before it runs and an exact receipt after. See Credits & Billing for the full breakdown.