# Opencanvas > An infinite canvas your local LLM draws on. Typed widgets, BYO model (Claude/OpenAI/Ollama), MCP-native, local-first, MIT. - **Type:** MCP server - **Install:** `agentstack add mcp-ashark-ai-05-opencanvas` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [ashark-ai-05](https://agentstack.voostack.com/s/ashark-ai-05) - **Installs:** 0 - **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [ashark-ai-05](https://github.com/ashark-ai-05) - **Source:** https://github.com/ashark-ai-05/opencanvas - **Website:** https://github.com/ashark-ai-05/opencanvas#readme ## Install ```sh agentstack add mcp-ashark-ai-05-opencanvas ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # OpenCanvas ### Stop reading AI replies. Watch them assemble on an infinite canvas. Your LLM doesn't print markdown — it **places typed widgets** on a tldraw canvas: charts, diagrams, code, kanbans, tables, custom HTML, anything. The agent **learns your patterns** and registers reusable widget templates over time. Works with any model that supports function calling — BYO key. **BYO model · MCP-native · Self-improving · Local-first · MIT** [](https://github.com/ashark-ai-05/opencanvas/actions/workflows/ci.yml) []() []() [](./LICENSE) [](./SECURITY.md) ▶︎ Try the demo  ·  ⭐ Star on GitHub  ·  📡 API reference  ·  🏗 Architecture **[Install](#install) · [Why](#why-its-different) · [Pick a model](#pick-any-llm) · [REST API](#drive-it-from-any-process) · [Security](./SECURITY.md)** --- ## What makes it different | | OpenCanvas | Claude Artifacts | Cursor Composer | v0 (Vercel) | ChatGPT Canvas | |---|---|---|---|---|---| | **Canvas-spatial UI** | ✅ tldraw infinite canvas | ❌ single artifact pane | ❌ single editor pane | ❌ single preview pane | ⚠️ document only | | **Multiple typed widgets per turn** | ✅ markdown · chart · kanban · code · mermaid · custom HTML · … | ⚠️ one at a time | ❌ | ⚠️ one UI at a time | ❌ | | **BYO model** | ✅ any provider (Anthropic / OpenAI / Google / Groq / Ollama / OpenRouter) | ❌ Claude only | ⚠️ OpenAI + Anthropic + a few | ❌ Vercel-hosted | ❌ OpenAI only | | **MCP-native** | ✅ first-class | ✅ via Claude.ai | ⚠️ recent | ❌ | ❌ | | **Self-improving widgets** | ✅ agent extracts templates from its own renders | ❌ | ❌ | ❌ | ❌ | | **Self-hosted / open source** | ✅ MIT | ❌ | ❌ | ❌ | ❌ | | **Local-first option** | ✅ runs entirely on your machine | ❌ | ⚠️ partial | ❌ | ❌ | --- ## Install ```bash git clone https://github.com/ashark-ai-05/opencanvas.git cd opencanvas pnpm install cp .env.example .env # at least one provider key pnpm electron:dev # backend + Vite + Electron, one command ``` Headless without Electron: `pnpm dev` → http://127.0.0.1:3458 > Packaged installers (`.dmg` / `.exe` / `.AppImage`) ship via the [Releases tab](https://github.com/ashark-ai-05/opencanvas/releases). Want to host a public demo? See [docs/deploy-railway.md](./docs/deploy-railway.md) — Railway + Groq Llama 3.1 70B, ~10 minutes. --- ## Why it's different | | Most chat apps | **OpenCanvas** | |---|---|---| | **Output** | A long stream of text | Typed widgets on a canvas — click, drag, pin, link, export | | **Memory** | Per-message context window | Self-improving SQLite KB; every chat is searchable across conversations | | **Tools** | A handful, baked in | 15 widget kinds + 12 agent tools + your MCP servers + runtime plugins | | **Sources** | Chat-only | MCP-native — any server (Confluence, Jira, filesystem, GitHub, …) | | **Drive** | Chat input only | REST API — any local process can render widgets via `POST /v1/canvas/*` | | **Privacy** | Cloud round-trips | Single-user, BYO credentials, runs entirely on your machine | --- ## Pick any LLM Every provider gets the **same agent** — same tools, same widget surface, same UX. The model is a config switch: | Provider | Auth | Notes | |---|---|---| | **Anthropic** | `ANTHROPIC_API_KEY` | Claude Sonnet/Opus 4+; extended thinking auto-on | | **OpenAI** | `OPENAI_API_KEY` | GPT-4o, GPT-4.1, o-series | | **Google Gemini** | `GOOGLE_API_KEY` | Gemini 2.5/3.x Flash & Pro; thinking surfaces in the chat | | **Groq** | `GROQ_API_KEY` | Llama / Mixtral / Kimi at Groq speeds | | **OpenRouter** | `OPENROUTER_API_KEY` | One key, hundreds of models | | **Ollama** | _none — local_ | Any model you've pulled (`llama3.2`, `qwen2.5-coder`, …) | | **Sourcegraph Amp** | `AMP_API_KEY` | Hosted agent loop (legacy path) | Tool calling is driven by the [Vercel AI SDK](https://ai-sdk.dev/), so any provider with function-calling support works the same way. See [`docs/plans/unified-agent.md`](./docs/plans/unified-agent.md) for the architecture. Embedders are pluggable too: bundled ONNX (runs offline), OpenAI, Voyage, or Ollama. Fully-local config example ```jsonc // ~/.opencanvas/config.json { "activeProfile": "local", "profiles": [{ "name": "local", "llm": { "provider": "ollama", "model": "llama3.1:8b" }, "embed": { "provider": "ollama", "model": "nomic-embed-text" }, "sources": [] }] } ``` --- ## The widget catalog **15 kinds.** Markdown · code blocks · tables · timelines · file trees · kanban · tasks · sticky notes · composite cards · charts (Vega-Lite) · calendars · clock/timer/stopwatch/pomodoro · web embeds (sandboxed iframes) · plugin-rendered iframes · generic fallback. Every widget is a **typed contract** — Zod schema for props + a React component. Add a new kind without retraining anything. **Runtime plugins.** The agent can call `register_widget_kind` mid-conversation to declare a new widget on the fly. Ships with a Python REPL (Pyodide) and a JS REPL example. Plugin iframes run with `sandbox="allow-scripts"`, no `allow-same-origin` — null origin, no parent DOM access. --- ## MCP-native Add any MCP server under `profiles[].sources` — works with any LLM provider: ```jsonc { "sources": [{ "id": "dev-fs", "transport": "stdio", "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/me/Development"] }] } ``` `pnpm cli --probe-sources` to verify. The agent invokes them as `mcp__dev-fs__`. --- ## Drive it from any process Any local process — cron jobs, watchers, scripts — can render widgets. The browser keeps an SSE connection open to `/v1/canvas/events`; external `POST`s push directives into the per-conversation event bus. ```bash # Place a Vega-Lite chart from your terminal curl -X POST http://127.0.0.1:3457/v1/canvas/widgets \ -H 'content-type: application/json' \ -H "x-opencanvas-token: $(cat ~/.opencanvas/auth-token)" \ -d '{"kind":"chart","role":"primary","payload":{"title":"Sales","spec":{...}}}' ``` **Full REST reference**: browse the [live API docs](https://ashark-ai-05.github.io/opencanvas/api.html) (Scalar / Swagger UI), or read the raw [OpenAPI 3.1 spec](./docs/openapi.yaml). When the backend is running, the same UI is served at `http://127.0.0.1:3457/docs`. **Auth.** A 256-bit token is generated on first run and persisted to `~/.opencanvas/auth-token` (mode `0600`). Required on state-mutating routes. See [SECURITY.md](./SECURITY.md). --- ## Self-improving KB Every conversation auto-indexes into the same SQLite + sqlite-vec store as your docs and code. Old turns become searchable via `search_kb`. Hybrid BM25 + vector retrieval. Idempotent re-indexing — re-running on unchanged content costs zero LLM calls. ```bash pnpm cli --index ./docs # markdown / text → chunks + embeddings pnpm cli --index-code ./src # tree-sitter chunks + symbols pnpm cli --search "" # hybrid search ``` The agent also learns *you* — kinds you keep + pin score higher; kinds you dismiss bias the next placement away. Per-conversation, persisted locally, no remote tracking. --- ## Architecture ``` ┌────────────────────┐ ┌─────────────────────────┐ │ Vite + React + │ /v1/chat │ Hono backend │ │ tldraw │ ─────────────→ │ │ │ • floating chat │ │ 6 LLM adapters │ │ • canvas │ │ 12 in-process tools │ │ • palette ⌘K │ ←── SSE ────── │ + external MCP servers │ │ • history │ /v1/canvas/ │ + plugin registry │ └────────────────────┘ events │ │ │ │ ▼ │ │ SQLite + sqlite-vec │ └─────────────────────────┘ ``` tldraw v3 with one shape util per widget kind. Zustand for app state. AI SDK 6 for chat streaming. Hono + better-sqlite3 + sqlite-vec on the backend. Electron wrapper for the desktop install. --- ## Development ```bash pnpm test # backend (vitest, Node) pnpm vitest run --config app/vite.config.ts # frontend (vitest + jsdom) pnpm typecheck # tsc --noEmit pnpm app:build # production bundle pnpm dist # electron-builder installer ``` **596 tests passing** · 383 backend, 213 frontend. --- ## Status Experimental. Solo project, MIT. No telemetry, no remote services unless you point it at one. Issues + PRs welcome. See [CONTRIBUTING.md](./CONTRIBUTING.md). For security issues, [SECURITY.md](./SECURITY.md). ## Source & license This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [ashark-ai-05](https://github.com/ashark-ai-05) - **Source:** [ashark-ai-05/opencanvas](https://github.com/ashark-ai-05/opencanvas) - **License:** MIT - **Homepage:** https://github.com/ashark-ai-05/opencanvas#readme Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** yes - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** yes - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/mcp-ashark-ai-05-opencanvas - Seller: https://agentstack.voostack.com/s/ashark-ai-05 - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.