FetchSandbox

About

FetchSandbox combines a locally launched MCP server with a hosted sandbox backend for testing API integrations from AI coding tools. An agent can import an OpenAPI 3.x document from a public URL or pasted JSON/YAML, receive a sandbox endpoint, inspect available workflows, and execute multi-step scenarios with schema-shaped responses and retained state.

The public GitHub documentation explicitly describes three core MCP tools: import_spec, list_workflows, and run_workflow. Imported specifications can match curated provider configurations or fall back to generated create-and-verify workflows for detected resources.

Key Features

• Imports OpenAPI 3.x specifications from URLs or pasted content.
• Creates stateful sandbox environments that preserve resources and lifecycle transitions between workflow steps.
• Runs curated or inferred workflows and returns step-by-step request and response traces.
• Supports webhook-oriented validation, including provider-style events and shareable receipt URLs for completed runs.
• Includes curated integrations for providers such as Stripe, GitHub, Twilio, OpenAI, Paddle, Clerk, and Resend.
• Offers an optional skills layer that can inspect a repository, route an integration task, request user approval before changes, and produce a structured ship recap.

Use Cases

FetchSandbox is suited to developers who want an MCP-capable agent to explore an unfamiliar API, verify a payment or authentication flow, test webhook-dependent behavior, or reproduce integration failures without consuming real provider quota. It can also help teams review brownfield integrations by generating workflow traces and public run receipts that can be attached to pull requests or shared with collaborators.

Compatibility

The official repository documents stdio configurations for Claude Desktop, Claude Code, Cursor, Cline in VS Code, Continue.dev, Codex CLI, and Zed. The official install guide also lists Windsurf. Generic stdio setup is documented for other MCP-compatible clients such as Roo and Goose. Lovable support is not documented in the official sources checked.

Limitations

FetchSandbox uses its hosted backend by default. The MCP package and CLI are MIT-licensed, but sandbox execution limits, private integrations, team controls, and enterprise deployment options belong to the hosted service.

For OpenAPI specifications without a curated configuration, generated workflows may use generic UUIDs and enum values rather than provider-specific identifiers or behavior. These runs are useful for request and response shape validation but may not reproduce every vendor-specific lifecycle or failure mode.

The current install page says a successful Cursor connection displays eight tools, while the public GitHub README enumerates only three core tools. The additional tool names and schemas were not publicly enumerated in the static official documentation checked, so this entry lists only the tools that could be verified individually.

Anonymous usage telemetry is enabled by default in the documented package configuration. It can be disabled with the documented telemetry environment variable.

Sources

• Official website: https://fetchsandbox.com/
• Install guide: https://fetchsandbox.com/install
• Workflow guide: https://fetchsandbox.com/how-it-works
• GitHub repository: https://github.com/fetchsandbox/mcp
• npm package: https://www.npmjs.com/package/fetchsandbox-mcp
• Pricing: https://fetchsandbox.com/pricing

MCP Setup

Prerequisites

• Node.js 18 or later
• npm and npx available
• An MCP-compatible client
• Internet access to the hosted FetchSandbox backend unless using a separately documented self-hosted deployment
• The documented client configurations use npx and do not require a global install; the website also offers npm i -g fetchsandbox-mcp as an optional package-install step

Access requirements

• No API-provider API key, OAuth flow, partner onboarding, or production credential is required for bundled sandbox runs
• The hosted free tier is documented as 1,000 runs per month with no payment card required
• Higher run limits, private integrations, team controls, and some deployment options require paid service tiers

Companion app or plugin

Name: FetchSandbox skills bundle Install URL: https://fetchsandbox.com/skills/install.sh

• For Cursor or Claude Code, run curl -sL https://fetchsandbox.com/skills/install.sh | bash if the ./fetchsandbox trigger phrase does not load.

Optional skill files for the guided FetchSandbox workflow; they are not required to launch the core stdio MCP server.

Client setup

Claude Code

Transport: stdio

npx -y fetchsandbox-mcp@latest

{
  "mcpServers": {
    "fetchsandbox": {
      "command": "npx",
      "args": ["-y", "fetchsandbox-mcp@latest"]
    }
  }
}

• Add the configuration to ~/.claude/settings.json for user-level use or .mcp.json in the repository root for project-level use.
• Restart the Claude Code session.

The optional skills bundle is documented for Cursor and Claude Code when the guided trigger phrase is not loaded.

Claude Desktop

Transport: stdio

npx -y fetchsandbox-mcp@latest

{
  "mcpServers": {
    "fetchsandbox": {
      "command": "npx",
      "args": ["-y", "fetchsandbox-mcp@latest"]
    }
  }
}

• On macOS, add the configuration to ~/Library/Application Support/Claude/claude_desktop_config.json; on Windows, use %APPDATA%\Claude\claude_desktop_config.json.
• Quit Claude Desktop completely and reopen it.

Cursor

Transport: stdio

npx -y fetchsandbox-mcp@latest

{
  "mcpServers": {
    "fetchsandbox": {
      "command": "npx",
      "args": ["-y", "fetchsandbox-mcp@latest"]
    }
  }
}

• Add the configuration to ~/.cursor/mcp.json for global use or .cursor/mcp.json for a project.
• Quit Cursor completely and reopen it.
• Open Settings, find MCP Servers, and toggle fetchsandbox on.
• Verify that the server shows a green status indicator.

The current official install page says Cursor should show eight tools, although the public GitHub README individually documents only three core tools.

Codex

Transport: stdio

npx -y fetchsandbox-mcp@latest

[mcp_servers.fetchsandbox]
command = "npx"
args = ["-y", "fetchsandbox-mcp@latest"]

• Add the TOML configuration to ~/.codex/config.toml.
• Restart Codex CLI.

Documented for OpenAI Codex CLI.

Windsurf

Transport: stdio

npx -y fetchsandbox-mcp@latest

• Use the Windsurf tab in the official FetchSandbox install guide to add the stdio server, then restart Windsurf.

Windsurf is listed as supported by the official install guide, but the exact Windsurf config path and block were not exposed in the static official page text checked.

VS Code

Transport: stdio

npx -y fetchsandbox-mcp@latest

• In the Cline extension, open the Cline panel, select the settings cog, open MCP Servers, and add a server with command npx and arguments -y fetchsandbox-mcp@latest.
• Reload the VS Code window.

The official repository documents VS Code setup through the Cline extension; the product website also lists VS Code compatibility.

Lovable

Lovable support and an exact public setup are not documented in the official sources checked.

Other

Transport: stdio

npx -y fetchsandbox-mcp@latest

• In any MCP-compatible client that accepts a stdio command, register npx with arguments -y and fetchsandbox-mcp@latest.
• Restart or reload the client after saving the configuration.

Official documentation includes Continue.dev and Zed examples and gives generic stdio guidance for clients such as Roo and Goose. The website install guide also lists Cline.

Authorization flow

• No API-provider OAuth or API-key authorization step is documented for bundled sandbox workflows
• No browser sign-in step is documented for launching the core MCP server

Environment variables

• FETCHSANDBOX_BASE_URL - Overrides the hosted backend URL for staging tests or a separately available self-hosted backend.; https://fetchsandbox.com
• FETCHSANDBOX_TELEMETRY - Set to 0 to disable the anonymous usage telemetry described in the official GitHub README.; on

Setup docs: https://fetchsandbox.com/install