Presentation Intelligence のリモート MCP サーバーとは何ですか?
Presentation Intelligence のリモート MCP サーバーは、https://mcp.pi.inc/api/v1/mcp にホストされた Model Context Protocol エンドポイントです。AI アシスタントはローカルに何もインストール・実行することなく接続できます。

@Pi Inc.
基本情報
Presentation Intelligence is an AI-native presentation creation and design platform. Through this remote MCP server, AI assistants can automatically generate, structure, and refine professional slides via natural language.
https://mcp.pi.inc/api/v1/mcpclaude mcp add presentation-intelligence --transport http https://mcp.pi.inc/api/v1/mcpRead account, credits, admission and pricing facts without creating or charging anything. Optional for creation: call it only when the user asks about account/pricing or when its facts are needed. Requires pi:account:read; an older grant may need host-managed consent for this scope, not reinstalling or changing accounts. This snapshot is not a final quote or user authorization. Never promise an unsupported spending cap. Parent-document pricing excludes later media charges; do not estimate those from pages, images, or themes. New creation may omit share_permission for the backend default of 1; a requested or default 1 requires pi:documents:share, while 0 only declines a new anonymous link and is not owner-exclusive. Retain auth_context internally for later calls. Theme-only and existing-task requests do not need this tool.
List existing themes, or search with preferences.query (1-300 characters) using a document title/topic or visual/template preferences such as warm colors or a spring launch style. Each theme may include preview_url and MCP-derived preview_markdown. When presenting previews, use preview_markdown verbatim if present; never emit empty-label links such as [ ](preview_url) or [](preview_url). If preview_markdown is absent, list name and theme_ref, provide the clickable preview_url when available, and say the preview is unavailable. This read is optional before creation: use it when the user selected a theme or asks for help choosing one. When asked to recommend from a title, pass that title as query and the selected PPT type as generation_mode. A title alone does not make theme search a mandatory creation preflight. Respect an explicitly selected theme; otherwise pi_create_document may omit theme_ref unchanged for one backend-owned automatic recommendation within the PPT category using the input title/topic before generation. That automatic recommendation uses the input title, not the final generated title; if it has no result, fails, or is unknown, it may use the frozen same-category default without another call or retry. Natural-language visual preferences do not guarantee overriding the backend default theme. New creation may omit share_permission for the backend default of 1, which requires pi:documents:share; explicit 0 only declines a new anonymous link and is not owner-exclusive. Optional color_id, style_id, scene_id, is_public filters still apply. preferences.generation_mode filters the PPT category: image-ppt uses Creative themes, svg-ppt uses Canvas themes, and ppt uses other primary styles. Without a mode, ordinary browsing spans categories and semantic search uses the backend image-ppt default. Category matching does not guarantee renderer availability or visual quality. Pass the same selected mode in settings.generation_mode when creating. Search requires offset=0 and returns at most 10 candidates in relevance order from a bounded pool, not guaranteed matches; disclose no_match/failure or an unsuitable preview and use alternatives only within authorization, never silently default. Does not create Generation, document or task resources or call document billing. Theme-selection permission alone does not authorize embedding-provider spending. Explicit semantic search may incur embedding-provider cost; it is not guaranteed free or repeat-safe, and must not be automatically retried or paginated. Preserve auth_context on later calls.
Generate one PI-native presentation document (PPT, PowerPoint, slides, or 演示文稿), potentially consuming credits. Use this only for an explicit request to generate when provider selection chooses PI: the user names or selects PI, a known PI default applies with no provider named in this request, or, with no default, host-visible capabilities and rules clearly select PI, including when PI is the only suitable visible provider. If the user names another provider, respect it regardless of a previous PI default. Ask one brief choice question only when no explicit provider and no clear default leave multiple suitable visible providers plausible. Do not call this tool when a strict .pptx file is a required delivery condition and PI has no such export: state that limitation first, and create only after the user explicitly accepts a PI online document instead. Installation, OAuth, discovery, or routing keywords do not establish a default or authorize a paid call. Read-only analysis, theme, outline, and existing-task requests do not use this tool. A generic PPT request is not a strict .pptx requirement. A user's explicit request to generate authorizes the existing backend charging rules and the disclosed default share_permission=1; do not add a second cost-confirmation or sharing-confirmation step. Requires pi:documents:create and pi:tasks:read before dispatch. share_permission accepts only integer 0 or 1, not booleans or null: omit it for the backend default 1, which requires pi:documents:share; 0 declines a new anonymous link but is not owner-exclusive. delivery.share_permission is the actual document ACL mask, while delivery.requested_share_permission is the frozen 0/1 request intent; neither missing field is inferred. Return a document link only after Backend confirms either strict ready delivery or verified succeeded+partial delivery. A partial link must include saved usable pages, Backend missing_media_count, actual ACL evidence, and a valid URL; present it as partial and never as a complete deck. theme_ref is optional: forward an explicit valid ref unchanged. If omitted, MCP forwards no theme_ref and the Backend performs one automatic recommendation within the PPT category using the input title/topic before generation; it is not a final generated title. A no-result, failure, or unknown recommendation may be recorded as fallback to the frozen same-category backend default without another recommendation call or retry. That backend recommendation may incur embedding-provider cost, is not promised free, and is not claimed in billing.total. Do not force pi_list_themes before creation. settings.generation_mode may be ppt, image-ppt, or svg-ppt; omit it for the Backend's image-ppt default, forward an explicit mode unchanged, and report an unsupported mode without fallback. visual_requirements is an optional 1-4000 character, non-whitespace user-constraint string for image-ppt; it may accompany an omitted mode so the Backend applies its default, but is rejected with explicit ppt or svg-ppt. It is validated without trimming, so the original supplied text is preserved. It is not inferred from a theme and does not promise that a theme's name, preview, or summary is the effective image style. Null is not a mode. SVG-PPT is an SVG-backed PI document, not an SVG or PPTX export. Keep user settings and backend defaults; The configured MCP default output language is "en"; if settings.language is omitted, null, or blank, MCP injects that value. An explicit non-empty settings.language takes precedence and is forwarded unchanged without MCP normalization. Use the user's requested output language when it is clear; do not equate the conversation language with the document language when the user says otherwise. Ask only when the output language is genuinely ambiguous and materially affects the result, not for every mixed-language request. Unknown language values may be forwarded but are not guaranteed to generate correctly. Never promise an unsupported …
Check an existing task directly; no status/theme preflight needed. Supply task_id OR the original idempotency_key, and retain expected_auth_context. Read-only, no generation or charge. Delivery must report actual share_permission; 0 is not owner-exclusive and absent ACL facts must not be invented. Follow guidance: From the original task reference, poll for at most 5 minutes or 60 GET attempts, whichever comes first; failed GETs count, request time and backoff count toward time, and progress or partial pages never reset either limit. Only poll when backend next_action=poll and poll_after_ms is positive: wait max(5 seconds, poll_after_ms). Stop on the 3rd consecutive or 6th cumulative retryable query failure; use 5 then 10 seconds for the first two consecutive failures and honor a longer Retry-After. A successful query resets only the consecutive-failure count. Auth or permission errors stop rather than making the task failed. Do not promise a background notification or that stopping queries pauses work. Stop immediately for terminal/blocked/awaiting_credits, poll_after_ms=0, stopped/manual-review recovery or exhausted budget. A query error is not task failure. Final document links are returned only with Backend-confirmed delivery: ready requires status=succeeded, next_action=deliver, delivery.status=ready, complete saved content, saved_page_count=total_page_count, and an actual valid view_url; verified partial requires status=succeeded, next_action=deliver, delivery.status=partial, saved_page_count>0, missing_media_count>0, complete ACL evidence, and a valid view_url. No URL is constructed from an ID. Present a partial result as readable but incomplete, with its missing-media explanation, and never rename it ready or claim all pages/media succeeded. Ignore any legacy early document_entry. A private or login-required delivered link is not automatic browser-open authority; do not change sharing, identities, or generate a replacement. Only ready delivery is eligible for automatic host embedding; partial requires explicit user-facing link delivery. Prioritize confirmed delivery. When billing.total is complete and valid, report it as the final bill and do not also present billing.observed as a separate cost; otherwise report only verified billing.observed as of this read, not as final settlement, and do not echo an incomplete or malformed total.
Read a persisted outline and version. This tool does not update the outline or create a document.
Save an outline draft using base_version and idempotency_key. Requires pi:outlines:write and pi:outlines:read before dispatch. This does not regenerate a document; read the persisted outline after an uncertain outcome.
Presentation Intelligence is an AI-native presentation creation and design platform[cite: 1]. With this remote MCP server, you can seamlessly connect AI assistants (like Claude and Cursor) to generate, structure, and design professional presentations through natural language.
Presentation Intelligence のリモート MCP サーバーは、https://mcp.pi.inc/api/v1/mcp にホストされた Model Context Protocol エンドポイントです。AI アシスタントはローカルに何もインストール・実行することなく接続できます。
エンドポイント https://mcp.pi.inc/api/v1/mcp を Claude Code、Cursor、VS Code など任意の MCP 対応クライアントに追加してください。このページのセットアップ手順で各クライアントを一度に設定できます。
はい。Presentation Intelligence は OAuth を使用します。初回接続時に MCP クライアントがブラウザを開いてサインイン・認可を行い、以降はその認証情報を再利用します。
Presentation Intelligence は streamable-http エンドポイントを公開しており、これはリモート MCP サーバーで広く使われ、主要な MCP クライアントがサポートするトランスポートです。
AOI is a commercial environmental-intelligence gateway that gives AI agents programmatic access to 38 real-world intelligence capabilities — climate risk, satellite monitoring, methane detection, water & soil, vegetation
Live crypto trading signals via API. Full trade setups with entry, stop-loss, take-profit, leverage, and automated verification. Built for Claude Code, OpenAI Codex, and 30+ AI agents.
Pay-per-use data and tools for AI agents: crypto token safety and rug checks, web search with cited answers, page-to-Markdown, up-to-date library docs, package safety, company research and UK/EU open data. 160 tools, 20
Official 42min MCP server: let Claude, ChatGPT, Cursor or any MCP client manage your meeting scheduling. Check availability, find free slots, book, cancel and reschedule meetings, manage recurring series, secret booking
ARVX™ reads client relationship health using the science of why partnerships last. It models two people, factors in what's influencing each of them, and reads their relationship across five areas, with what moved since t
コメント