Privacy and security

Studio is often used with private, internal APIs, so it talks only to the servers you point it at. This page lists what it sends, what it keeps, and how the parts that listen on your computer are protected.

What Studio sendsCopy Anchor Link

• Your requests only. No telemetry, analytics or crash reporting.
• Spec0 only when signed in, and then only what you ask for, such as your organisation's APIs or creating a hosted mock. History, environments, secrets, collections and logs are never synced.
• Update checks on request, or on start if you turn that on. It's one request to github.com and sends nothing about your work.
• Agent requests, after you allow them. An agent using the local MCP server can send requests through Studio, but each new host waits for your answer, and you can block hosts outright. See Approving what agents do.
• Agent chat messages. What you type and attach in the agent chat goes to that agent and from there to its model provider, with secret values replaced by their {{name}}. The chat is off until you choose an agent, and the agent starts only when you send a message. Studio never sees your sign-in to the agent. An Anthropic API key for Claude Agent, if you save one, is kept in the credential store and given only to the Claude agent. See Privacy and safety.
• No requests from the reference view. The rendered reference is given the spec's text, never a URL, and the window blocks its network requests. A remote image in a spec's description isn't loaded, because loading it would tell the spec's author that you opened it.

A Content Security Policy in Studio's window blocks any other network requests from the interface, whatever a bundled library tries to do.

What Studio keeps, and whereCopy Anchor Link

Everything is on your computer. Settings → Data & privacy shows the folder.

SecretsCopy Anchor Link

• One place for secrets. Auth values aren't stored per API. They go in an environment, where you mark them secret.
• The credential store. Secret values go in the macOS Keychain, Windows Credential Manager, or the Secret Service on Linux. See Where secrets are kept.
• Never written down. History, error messages, saved unsent changes, collection files, run logs, the console, agent activity, saved chats and exported reports show {{name}} instead of the value. An auth value typed straight into a request is kept only until Studio quits.
• Safe to commit. Environment files and collection files note that a secret exists, not its value.

CertificatesCopy Anchor Link

Certificate checks can't be turned off for everything at once. You can trust a private CA, or skip checks, for one host at a time, and Studio marks requests to that host as unverified. See Network settings.

Links from web pagesCopy Anchor Link

A spec0:// link can be sent by any web page, so Studio asks before doing anything. It shows the API's name, the host it would download from and the full address, and downloads nothing until you press Open. It only accepts https addresses on public hosts. See Open in Studio links.

Servers Studio runs on your computerCopy Anchor Link

Studio can run two kinds of server. Both are off until you start them, listen on 127.0.0.1 only (other computers can't reach them), and stop when Studio quits. Starting a chat in the agent chat starts the MCP server too, unless you turn that off in Settings → Agent chat.

See Local MCP server and Local mock servers.

Reporting a security problemCopy Anchor Link

Please report security problems privately, as described in the repository's security policy, rather than in a public issue.