MCP Confirmation Questions

Please answer these before the MCP docs move from draft to implementation-ready reference.

Confirmed So Far

1. Endpoint And Transport

1. Is https://dev.deckflow.com/mcp/v1/ the public production endpoint or a dev/beta endpoint?
2. Are there staging, regional, or additional versioned endpoints?
3. Should examples always include the trailing slash?
4. Is DeckFlow remote HTTP MCP only, or will there also be SSE, stdio, or local server support?
5. Should the canonical MCP config key be deckflow, DeckFlow, or another name?

2. Authentication

1. Where do users create or copy a DeckFlow API key?
2. What header, secret field, or environment variable should clients use for the API key?
3. Should docs standardize on an env var such as DECKFLOW_API_KEY, or use client-specific secret configuration only?
4. Are API keys personal, workspace-scoped, organization-scoped, or project-scoped?
5. How do users revoke or rotate a DeckFlow API key?

3. Official Client Support

1. Confirm that launch docs should include only Claude Code, Claude Web/Desktop, Manus, and OpenAI/ChatGPT.
2. Should Cursor receive a dedicated page later even though HeyGen only mentions it in overview?
3. Do any clients require domain whitelisting, app registration, or private beta access?
4. Which client pages need screenshots before publishing?

4. Tool Naming

1. Are create_deck, revamp_deck, and translate_deck acceptable final tool names?
2. Should Tools use grouped tools like convert_file, extract_file_content, compose_files, and optimize_file, or one tool per browser tool?
3. What are the final task tools: get_task, list_tasks, cancel_task, and get_task_result, or different names?
4. Is get_deckflow_capabilities the preferred discovery tool name?
5. Should there be get_current_user, list_brand_profiles, list_templates, list_glossaries, and list_supported_languages tools?

5. File Input And Output

1. How should agents pass files: MCP attachments, DeckFlow upload IDs, pre-signed upload URLs, external URLs, inline content, or a mix?
2. Can agents pass local file paths from terminal clients, or must they upload first?
3. How long are input files retained?
4. Temporary download links are confirmed. What is the signed URL lifetime?
5. Are outputs also persisted in a DeckFlow workspace, or only available through temporary links?

6. Task Lifecycle

1. What is the task status enum?
2. Does every task return progress percentage, stage name, ETA, or only status?
3. Which tasks can be cancelled?
4. Are idempotency keys supported to prevent duplicate submissions?
5. What is the standard error schema?

7. Product Workflow Parameters

1. What are the final Create input source types and output formats?
2. What Create parameters should exist for slide count, aspect ratio, language, Brand DNA, template, and complex element preservation?
3. What Revamp parameters should exist for style mode, template, Brand DNA, content preservation, batch mode, and output format?
4. What Translate parameters should exist for target language, source language, glossary, layout preservation, same-format output, and image text translation?
5. Which parameters are required, optional, or defaulted by DeckFlow?

8. Limits And Billing

1. Do MCP workflows use the same file limits as the public browser tools?
2. What are the maximum files per task, total batch size, slide count, page count, video duration, and inline prompt length?
3. What are the rate limits and concurrent task limits?
4. Which DeckFlow plans include MCP access?
5. How are credits consumed for completed, failed, cancelled, or partially successful tasks?

9. Security And Compliance

1. Should the docs state AES-256 encryption and post-processing deletion for MCP uploads too?
2. Are MCP inputs used for model training?
3. Are there workspace audit logs for MCP task creation and downloads?
4. Are there admin controls for enabling/disabling MCP?
5. Are any file types or domains blocked for security reasons?

10. Publishing

1. Which placeholders can be replaced now?
2. Which pages should remain internal until integrations are live?
3. Should examples include real screenshots, or text-only setup first?
4. Should docs use OpenAI, ChatGPT, or OpenAI / ChatGPT as the page title?
5. Who signs off on endpoint, billing, and retention language?