TinyScale

What TinyScale MCP server does

TinyScale MCP server gives a coding agent a way to configure hosted SaaS features for an application and verify that the integration works. Supported products include TinyWaitlist for consent-based launch signups, TinyFeedback for anonymous plain-text feedback, TinyChangelog for approved release notes, TinyCheck for browser-based deployment checks, TinyActivate for onboarding checklists, TinyImport for approved CSV data imports, TinyDemo for reviewed page walkthroughs, and TinyCite for reports on how AI answers describe a product.

The workflow is intended for setup tasks that can be completed from an existing project. For example, an agent can be asked to add a launch waitlist, add a feedback widget and changelog, or check that a deployed page displays the expected content. Public-facing actions such as accepting real email addresses or publishing release notes require approval.

How it works

The local MCP process runs in the project directory. During preview, the agent creates a project that remains available for 72 hours without signup or a payment card. Preview limits are small and stop when reached; the material does not describe preview usage as billable.

After setup, the agent is expected to demonstrate a real result. Depending on the product, that can be a feedback submission ID, a synthetic waitlist signup, or a passing browser check with a receipt. The user then opens a claim page and signs in. Claiming preserves the project, identifiers, and embeds on the Free plan.

TinyScale MCP server communicates with https://api.tinyscale.io; plain HTTP is used only for localhost development. Widgets run only on configured exact HTTPS origins, and browser checks require proof that the user owns the origin.

Setup and configuration

Install Node.js 20 or later for command-line use. Add the MCP process to Claude Code with:

claude mcp add tinyscale -- npx -y tinyscale mcp

Other clients can use a standard MCP configuration with npx as the command and -y tinyscale mcp as the arguments. The README specifically provides this configuration for Cursor, Windsurf, VS Code, and Codex. A Claude Code plugin is also available; it adds the MCP server and a skill describing the setup flow.

The process stores credentials in .tinyscale/agent-state.json with file mode 0600. The .tinyscale/ directory is git-ignored, and credentials and claim links are excluded from tool results. Configuration for individual products is supplied through the agent workflow or the corresponding product guide.

Tools and capabilities

TinyScale MCP server supports product discovery and workspace operations alongside product setup. The command-line examples show workspace discovery, workspace creation with selected products, waitlist configuration, project claiming, and general help. Product-specific guides are available for TinyWaitlist, TinyChangelog, TinyActivate, TinyImport, TinyDemo, and TinyCite; TinyFeedback and TinyCheck are also listed as supported products.

The server is a fit for agents that need to make a change, connect the relevant embed or check, and confirm the outcome rather than only generate integration code. It does not replace the claim step: the user must open the claim page and sign in to retain the project after preview.

Limitations and notes

Preview projects last 72 hours and have hard limits. The supplied material does not specify the numerical limits for each product. The service is hosted separately from the MIT-licensed client and has its own terms.

Security controls depend on the configured origin and local state handling. Widgets are restricted to exact HTTPS origins, while browser checks require ownership proof. The README does not list environment variables for configuration; local state is kept in the .tinyscale/ directory instead.