SandBase CLI

What SandBase CLI MCP server does

SandBase CLI MCP server gives an AI client one MCP interface for discovering and calling more than 2,000 AI models and APIs. The catalog can cover tasks such as language, image, video, search, social, and real-time data workflows, although the specific available endpoints and requirements should be checked before execution.

The server exposes discovery, inspection, execution, asynchronous status checks, run history, and account information. This makes it suitable for agents that need to select an endpoint dynamically rather than using one fixed provider integration.

How it works

The CLI connects an AI client to SandBase through a local MCP bridge. After the initial browser authorization, a supported client can use the MCP tools to work with the available catalog. The README describes the workflow as:

1. Search for candidate models or APIs with sandbase_discover.
2. Inspect the selected endpoint with sandbase_inspect to see its current input schema, pricing, and execution details.
3. Confirm the endpoint and parameters, then call it with sandbase_run.
4. For an asynchronous request, use the returned run ID with sandbase_run_get instead of submitting a duplicate request.
5. Review recent activity and recorded costs with sandbase_runs.
6. Check the account balance with sandbase_account.

This sequence lets an agent inspect requirements and possible cost before making a call.

Setup and configuration

The documented release command runs the CLI directly with npx and starts its MCP connection mode:

npx -y https://github.com/sandbaseai/cli/releases/download/v0.1.17/sandbaseai-cli-0.1.17.tgz connect

On macOS or Linux, the README also documents Homebrew installation followed by sandbase connect:

brew install sandbaseai/tap/sandbaseai-cli
sandbase connect

A browser authorization must be approved once. The project states that users do not need to manage individual API keys. For reproducible installation, the v0.1.17 archive and its published SHA-256 checksum can be downloaded and verified before execution.

The release also provides a read-only client compatibility check through catalog --json, which can be run before signing in or changing configuration.

Tools and capabilities

• sandbase_discover searches the available model and API catalog.
• sandbase_inspect returns an endpoint's input schema, pricing, and a ready-to-use template.
• sandbase_run executes a selected model or API endpoint.
• sandbase_run_get retrieves the status or result of an asynchronous run.
• sandbase_runs lists recent calls with status and cost information.
• sandbase_account checks the account balance without charging for the balance lookup.

Limitations and notes

The available endpoint catalog, schemas, pricing, and execution requirements should be treated as changeable and inspected at runtime. Calls can incur costs, so agents should inspect pricing before execution and use run polling for asynchronous work. The README states that the npm latest tag may not match the v0.1.17 GitHub release, so users needing that exact version should use the immutable release archive or the documented Homebrew route. Client compatibility should be checked with the provided catalog command rather than assumed from the presence of MCP support.