Stytch

No. Anthropic's directory sets this connector's permissions field to Read, but eleven of the twenty-one tool names in that same listing begin with create, update, delete or set. The directory's own example prompts ask Claude to create a project and configure redirect URLs, so the listing contradicts itself within a single entry.

Yes. Stytch's OpenAPI specification for the create endpoint states the response contains the full secret value, which will not be exposed in future Get requests, and its response schema marks that secret string as required. A live Stytch project credential therefore lands in the tool result and in your chat transcript.

Yes, if you grant the broad scopes. Stytch's authorization server advertises manage:api_keys alongside a separate manage:api_keys:test, so a test-only key permission exists as a distinct scope. Every Stytch project ships with one Live and one Test environment, and the tools address environments by slug rather than defaulting to Test.

Eight scopes. Its RFC 9728 descriptor lists openid, email and profile for identity, then admin:projects, manage:api_keys, manage:api_keys:test, manage:project_settings and manage:project_data for the workspace itself. Stytch documents that some MCP clients request everything in scopes_supported as their initial permission set, so review the consent screen.

Yes, if the application is still using it. Stytch's delete endpoint takes a secret ID and returns only a request ID and status code, with no confirmation step or restore path documented. Because reads return only the last four characters, an agent cannot reliably tell which secret an application currently depends on.

Yes. setPasswordStrengthConfig requires three booleans, including check_breach_on_creation and check_breach_on_authentication, and accepts a validation_policy of ZXCVBN or LUDS with a configurable minimum length. Sending a partial configuration can therefore switch off breached-password detection for real end users as a side effect of an unrelated edit.

Stytch caps programmatic creation at 50 projects per workspace, 50 secrets per project, 50 public tokens per project, 100 redirect URLs per project and 100 email templates per project. One event log streaming destination is allowed per project. These are absolute limits on objects that can be created programmatically, not plan tiers.

Anthropic's directory lists Claude and the Claude API only, not Claude Code or Claude Desktop. Stytch's own documentation walks through Cursor and offers one-click installation for Cursor and VS Code from mcp.stytch.dev. The endpoint is standard streamable HTTP, so any MCP client that completes an OAuth flow can connect.