Discover and execute third-party tools through a metered agent interface
Cracked is a tool-discovery and execution platform for AI agents. Instead of building a separate integration for every external service, a developer can discover tools, inspect their schemas and execute supported calls through a common interface. The offering includes MCP, HTTP access, SDKs and a command-line workflow.
The platform presents information about tool fit, health and price, alongside the parameters required for a call. An agent or developer should inspect those details before execution, especially when a tool sends data outside the application or changes an external record. Provider credentials can be supplied directly where supported, while other routes use the platform's account arrangement. Those choices affect both billing and the permissions involved.
The published pricing distinguishes ordinary calls, bring-your-own-key execution and waterfall attempts. Standard calls include a provider-cost markup and a small platform charge. Waterfall behavior has separate rules and can incur actual provider attempt costs even when no final hit is returned. It would therefore be misleading to say every unsuccessful workflow is always free. Cost inspection, limits and execution receipts are important parts of a production integration.
Cracked advertises an introductory workspace credit rather than a permanent free plan. Purchased-credit conditions and provider charges should be reviewed before a larger deployment. Its retrieved privacy page explicitly describes the legal text as a preview pending final review, which is a material limitation for organizations handling sensitive information. Begin with non-sensitive test inputs, validate result schemas and check how timeouts or asynchronous work are handled. The platform can simplify integration discovery, but each selected tool still needs an appropriate authorization and data-handling review.
