Use Kanvas Help with an agent
Version 0.20.0. Learn a user-chosen task with live context and real interaction.
1. Respond to the intent of a request for help, not a keyword trigger. Use kanvas_help read with a focused topic/question and optional authorized session/display context. A general request gets a short orientation and a few choices; a specific how-to gets the relevant recipe.
2. Do not publish anything merely because Help was opened. Once the user requests an activity, use normal publishing/control tools within that authorization without repeatedly asking to perform routine steps.
3. Use the agent host’s available search or domain tools only where the task benefits from them. Attribute supplied sources; if tools/data are unavailable, say what is missing instead of silently inventing live facts.
4. Branch with the user’s choices and existing session state. On reconnect, read current view and pending interactions; do not start a fixed sequence. Read and acknowledge actual interaction IDs once, and verify receiving-browser outcomes before describing success.
5. For a personalized walkthrough, read help topic tour or guide topic tour: use available authorized memory/profile and connected tools to construct a relevant live experience.
6. For “what is Kanvas?” or “show me what you can do”, compose a live answer from the current tool descriptions, template list and guides (kanvas_help guide, kanvas_template get, kanvas_help read) and publish real, relevant views. There is no fixed or pre-built demo deck.
7. MCP guidance cannot force host tool loading, idle-agent wakeups, narration style or auto-approval. User intent and existing permission remain authoritative.
Related tasks
- A tour built around your work — Agent instructions for a live, context-aware tour, never a fixed demo deck.
- Kanvas Help — Get a useful visual onto a screen, then work with it together.
- Publish your first view — One call sets up a fresh authorized workspace and returns real viewer links.
- Navigate, restore history and autoplay prepared views — Page, scroll, find, highlight and focus without republishing source data.
- Connect a private screen or open a public session — Invite a browser without confusing its address with permission.
- Compare products or offers — Line up vendors and ordered attributes with exact prices and complete URLs.
- Show external websites — Edge-to-edge public pages, optional mobile viewports and honest static fallbacks.
- Set numbers, units and dates — Use regional display preferences without changing exact source strings.
- Choose a template — Use the current catalog and exact field schema instead of guessing.
- Recover from a failed call — Find the real error, correct the input, and retry without duplicating work.
- Read comfortably at your own pace — Independent text size, zoom and words-per-minute controls for document reading.
- Show task progress and exact logs — Accumulate real milestones, inspect dense logs and use proportional timelines.
- Invite someone to KANVAS — Create a welcome link without granting private access or sending a message.
- A live Home for your sessions — A real KANVAS display, deliberate session choices and ordinary OAuth.
- Install KANVAS and choose a receiving device — An installable web app and account-registered browser receivers.
- Read messages and technical work — Source-driven email, inbox, chat, cards, bookmarks and developer/operations views.
- Filter, sort and highlight source records — Agent-authored presentation rules, without extra filter widgets.
- Profiles, organizations and independent teams — Optional photos, clear appearance precedence and explicit group access.
- Your activity history and access audit — Query actual retained events and create charts within your current permissions.
- Live people, devices, browser windows and participant map — A live connection hierarchy, distinct from the chronological overview list.
- Explore exact JSON as a tree — A structured explorer for nested source JSON, separate from code rendering.
- Point together at the same content — Optional named semantic markers that follow text and data across screen sizes.
- Show full images, team logos and drawing backgrounds — Source-preserving media with optional per-image information.
- Report an issue at the right scope — Separate session observations, private account feedback and general service requests.