Chatty in your AI client

A remote MCP endpoint to manage assistants, catalogs and conversations with an API key.

Connect your client

1. Open API / AI Integrations in the Chatty dashboard. A Pro or Business account is required.
2. Create an API key and copy it immediately: it is shown only once.
3. Configure a Streamable HTTP client supporting custom Bearer headers, such as Claude Code.

claude mcp add --transport http chatty https://api-chatty.romiltec.it/mcp \
  --header "Authorization: Bearer chatty_sk_..."

Replace the placeholder in your private configuration. Never publish the key in widget snippets.

The key retains your full account permissions. Revoke it from the API dashboard; OAuth and granular permissions remain phase 2. Connection through the ChatGPT and Claude.ai web interfaces has not been tested.

Choose your client

MCP connects the client, not the model: tools, Streamable HTTP and Bearer headers are required. These configurations follow vendor documentation; they do not certify every model or application. The key must be available to the process launching the client.

[mcp_servers.chatty]
url = "https://api-chatty.romiltec.it/mcp"
bearer_token_env_var = "CHATTY_API_KEY"
tool_timeout_sec = 180

{
  "mcpServers": {
    "chatty": {
      "url": "https://api-chatty.romiltec.it/mcp",
      "headers": {
        "Authorization": "Bearer ${env:CHATTY_API_KEY}"
      }
    }
  }
}

{
  "mcpServers": {
    "chatty": {
      "httpUrl": "https://api-chatty.romiltec.it/mcp",
      "headers": {
        "Authorization": "Bearer chatty_sk_..."
      },
      "timeout": 180000,
      "trust": false
    }
  }
}

{
  "mcpServers": {
    "chatty": {
      "type": "http",
      "url": "https://api-chatty.romiltec.it/mcp",
      "headers": {
        "Authorization": "Bearer chatty_sk_..."
      }
    }
  }
}

Verify the connection

Run list_assistants, then get_assistant with the selected UUID. Enable AI mode using update_assistant with body: {ai_mode_enabled: true}. After import_catalog, check list_catalog_items for status: ready. Test ask_assistant and reuse session_id for follow-up questions. Visual cards appear in the widget; the MCP client may show only text and JSON.

From catalog to answers

1. List your assistants with list_assistants and choose an instance.
2. Import products with import_catalog: inline CSV/JSON or Google Merchant XML/TSV, up to 10 MB and 10,000 rows. external_id updates existing products; products omitted from the file are retained.
3. Try a question with ask_assistant. This saves a conversation and consumes inference.
4. Read get_stats: recommended products, clicks and conversations help assess the answers.

The transport accepts 60 requests per minute per key. Remote tools do not read local paths or download URLs; imports wait for indexing.

Have PDFs or files on your computer?

The local .mcpb extension for Claude Desktop remains available in API / AI Integrations to upload files from disk.

The local 0.2.0 bundle uses list_chatbots and send_test_message and does not include catalog tools. Use the HTTP endpoint for catalogs.