Use Parrot in ChatGPT and Muse

How it works

1. In ChatGPT or Muse, add the Parrot connector.
2. Sign in to your Parrot account and review what the app can access.
3. Tap Allow. The assistant can now tailor Spanish practice to your level and your words.

Use another MCP app

You can also connect Parrot to an MCP app that runs on your own computer, such as a desktop AI assistant, a coding assistant, or the MCP Inspector. The app must support remote MCP servers with OAuth sign-in.

1. In your MCP app, add a remote MCP server (sometimes called Streamable HTTP) with the URL https://prod.parrotapp.com/mcp. No API key is needed.
2. The app opens your browser. Check that the page is on parrotapp.com, then sign in to Parrot there.
3. Check that the consent page says An MCP app on your computer wants to connect to your Parrot account, and review what it can access.
4. Tap Allow. Your browser sends you back to the app, and the Parrot tools appear.

Keep your connection safe

• Copy the server URL from this page. Do not use a Parrot server URL from another website or message.
• Type your Parrot password only on the Parrot page in your browser, never into the app or a chat.
• Parrot sends the sign-in result only to your own computer, so it cannot reach another website.
• Allow only apps you installed and trust. Every app you connect gets the access listed below.
• To disconnect, open Parrot and go to Settings → Connected Apps. Access ends right away.

AI apps that run on a website, not on your computer, can connect only after Parrot approves them. If your app shows an error that its redirect URI is not allowed, email support@parrotapp.com and tell us which app you use.

What the assistant can and cannot do

Parrot stays the source of truth. The connector never gives the AI app your Parrot password.

Example prompts

• “What's my Spanish level in Parrot?”
• “Quiz me on the words I'm learning in Parrot.”
• “Teach me five travel words at my level and save them to Parrot.”
• “Save mariposa (butterfly) as a word I'm learning.”
• “I practiced Spanish today, count it toward my streak.”

Privacy

Parrot shares only the data listed above, and only after you approve the connection. Access ends as soon as you disconnect. See the Connected AI Apps section of our privacy policy.

Support

Questions or problems? Email support@parrotapp.com.

Technical reference

For developers and reviewers

Server

MCP server URL:

https://prod.parrotapp.com/mcp

• Transport: Streamable HTTP, stateless JSON-RPC over POST /mcp. There are no sessions or server-sent event streams (GET and DELETE /mcp return 405).
• Supported MCP protocol versions: 2025-11-25, 2025-06-18, 2025-03-26, 2024-11-05.
• Methods: initialize, ping, tools/list, tools/call.

Authentication

• OAuth 2.1 authorization code flow with PKCE (S256 only). The learner signs in to Parrot and approves on a Parrot consent page.
• Scopes: learner:read (level, streak, word lists) and learner:write (save words, record practice).
• Discovery: an unauthenticated request to /mcp returns 401 with a WWW-Authenticate header pointing to the protected resource metadata.
• Client registration: supports Client ID Metadata Documents (the client_id is an HTTPS URL to the client's metadata) and Dynamic Client Registration (RFC 7591). All clients are public PKCE clients; redirect URIs must be HTTPS on a trusted AI app's domain, or loopback for local development tools.
• Local tools and scripts can skip registration and use the built-in public client ID parrot-mcp-local. It accepts any loopback redirect (http://127.0.0.1, http://[::1], or http://localhost, on any port) and still requires PKCE.
• Access tokens last 1 hour. Refresh tokens last 90 days and rotate on every use. Authorization responses include iss (RFC 9207).
• Revoking a token through the revoke endpoint makes the next call return 401.

Tools

• Every tool declares an input and output schema. Results include structuredContent.
• No tool is destructive. Read-only tools are annotated readOnlyHint: true.
• Tools that need learner:write return an error asking the user to reconnect with more access if the token only has learner:read.

Example tool result: get_learner_context

{
  "language": "es",
  "level": { "parrotLevel": 12, "cefr": "Lower B1" },
  "streak": { "dayStreak": 5 },
  "counts": { "knownWords": 830, "learningWords": 41 }
}

cefr is null if the learner hasn't taken the placement test.

Example tool input: set_word_status

{ "words": [{ "targetWord": "mariposa", "nativeWord": "butterfly", "status": "learning" }] }

Limits: targetWord up to 80 characters, nativeWord up to 200. Existing words keep their stored meaning and only change status.

Example tool result: list_learning_words

{
  "words": [{ "wordKey": "mariposa", "targetWord": "mariposa", "nativeWord": "butterfly", "status": "learning" }],
  "nextCursor": null,
  "totalCount": 41
}

Page size defaults to 100 (maximum 500). Pass nextCursor back as cursor for the next page; it is null on the last page.

HTTP API (same data, same JSON)

The same contract is available over REST with the same OAuth tokens, base URL https://prod.parrotapp.com:

Errors use this shape:

{ "error": { "code": "...", "message": "..." } }

Rate limits

120 requests per minute per connection (burst 30).