# Tapioca > Repositório central de skills, agents e MCPs reusáveis pro Claude Code e Cursor (estilo platform engineering). - **Type:** MCP server - **Install:** `agentstack add mcp-gabriel-dantas98-tapioca` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [gabriel-dantas98](https://agentstack.voostack.com/s/gabriel-dantas98) - **Installs:** 0 - **Category:** [Integrations](https://agentstack.voostack.com/c/integrations) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [gabriel-dantas98](https://github.com/gabriel-dantas98) - **Source:** https://github.com/gabriel-dantas98/tapioca - **Website:** https://gabriel-dantas98.github.io/tapioca ## Install ```sh agentstack add mcp-gabriel-dantas98-tapioca ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About tapioca We bring the base. You pick the filling. --- Distributable plugin for [Claude Code](https://code.claude.com/docs/en/plugins) and [Cursor](https://cursor.com/docs/reference/plugins). `tapioca` is a neutral base — you add fillings (skills + companion agents) under one namespace. PT — A nossa base, o seu recheio. Plugin com skills e agents sob o namespace `tapioca`. A marca é a base; cada skill é um recheio. Algumas fillings são de domínio PT-BR (ex.: `humanizer-br`); outras são agnósticas (ex.: `multi-gen`, `google-sheets-apps-script`). ## Included skills ### `/tapioca:multi-gen` Runs multiple AI CLIs in parallel (at least `codex` and `cursor-agent`) from an image briefing, validates each artifact (well-formed SVG), and builds a comparative HTML preview (light + dark + favicon strip at 16/32/48/180px) so you can pick a winner before manual polish. - Portable parallel dispatch (macOS without `gtimeout`) - Auth pre-check for `cursor-agent` (SKIP instead of hanging) - Robust `` extraction when CLI stdout mixes logs - Per-engine output: `raw.txt`, `out.`, `status.json` in the run dir ```text /tapioca:multi-gen "geometric logo for Novarum, blue palette" --palette "#2E5BFF,#FFFFFF" ``` Pairs with a `preview-server` skill (control-plane) to serve `index.html`. ### `/tapioca:usabilidade-br` Audits web app usability against Jakob Nielsen's 10 heuristics. Captures evidence via Chrome MCP, optionally correlates with source (`--code `), and writes a local self-contained HTML report with: - Overall score (0–100) and per-heuristic scores - Visual evidence (embedded screenshots) - Code snippets with `file:line` when the component is found - **Copyable fix prompt** per violation — paste into another Claude Code session Companion agent parallelizes 10 passes (one per heuristic) and builds the report. ```text /tapioca:usabilidade-br http://localhost:3000 --code ./src ``` ### `/tapioca:humanizer-br` Removes AI writing tells from Brazilian Portuguese and injects a human voice. Catalog of 25+ patterns: promotional tone, stacked gerunds, negative parallelism, rule of three, English-style Title Case, curly quotes, dramatic hooks, chatbot residue, and more. **Modes:** - Claude-only (default): runs in-session, zero extra cost. - Optional Maritaca: delegates rewrite to `sabia-3` (native PT-BR). Enabled when `MARITACA_API_KEY` is set. Companion: `humanizer-br` agent (under `agents/`) for multi-pass with self-scoring. ### `/tapioca:google-sheets-apps-script` Google Sheets and Google Docs automation via Apps Script (browser mode — no GCP for the end user). Docs are markdown-first. The agent runs the CLI; the user pastes links and confirms in Google's UI. ```text /tapioca:google-sheets-apps-script ``` Upstream: [gist v2.6](https://gist.github.com/gabriel-dantas98/6ad86b6bfab840703ec214f228c3004b). ## Install ### Claude Code ```bash git clone https://github.com/gabriel-dantas98/tapioca ~/.claude/plugins/tapioca ``` Restart Claude Code or run `/reload-plugins`. ### Cursor ```bash git clone https://github.com/gabriel-dantas98/tapioca ~/.cursor/plugins/tapioca ``` The repo ships manifests for both platforms (`.claude-plugin/plugin.json` and `.cursor-plugin/plugin.json`) over the same `skills/` and `agents/`. ### Via --plugin-dir (Claude Code, local dev) ```bash claude --plugin-dir /path/to/tapioca ``` ### Local smoke (marketplace + dual-CLI) Harness in `.agents/skills/smoke-test-skills/` — same source as CI: ```bash .agents/skills/smoke-test-skills/run.sh marketplace # token-free .agents/skills/smoke-test-skills/run.sh oneshot humanizer-br both # uses tokens .agents/skills/smoke-test-skills/run.sh all humanizer-br both ``` Cursor IDE: the harness creates `~/.cursor/plugins/local/tapioca` (symlink). Then: Developer → Reload Window. ## Usage ```text /tapioca:humanizer-br Paste the text to humanize here. ``` Or natural language: ```text Humanize this text, strip the AI smell. ``` For a serious multi-pass review, the companion agent is invoked automatically. ### Maritaca mode ```bash export MARITACA_API_KEY="your-key-here" ``` The skill detects the env var and offers Maritaca when the text justifies it (long enough, or quality explicitly requested). ## Layout ```text tapioca/ ├── .claude-plugin/plugin.json ├── .cursor-plugin/plugin.json ├── AGENTS.md ├── CLAUDE.md ├── README.md ├── skills/ │ ├── humanizer-br/ │ ├── usabilidade-br/ │ ├── multi-gen/ │ └── google-sheets-apps-script/ └── agents/ ├── humanizer-br.md └── usabilidade-br.md ``` ## Roadmap | Version | Content | |---|---| | **v0.1** | `humanizer-br` (skill + agent), plugin format, AGENTS.md | | **v0.2** | `usabilidade-br`, `multi-gen` | | **v0.3** | `google-sheets-apps-script`, marketplace manifests | | next | More fillings; official marketplace submissions | No hooks, MCP, or LSP in the short term. Skills and agents until the collection justifies more. ## Credit `humanizer-br` descends from: - [blader/humanizer](https://github.com/blader/humanizer) (MIT) — canonical English humanizer for Claude Code / OpenCode. - [mackswendhell/humanizer-pt-br](https://github.com/mackswendhell/humanizer-pt-br) (MIT, 2026) — first direct WikiProject AI Cleanup adaptation to PT-BR. What `tapioca` adds: namespaced plugin, Claude Code + Cursor, optional Maritaca `sabia-3`, multi-pass companion agents, and room for fillings beyond writing. ## License MIT — see `LICENSE`. ## Source & license This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [gabriel-dantas98](https://github.com/gabriel-dantas98) - **Source:** [gabriel-dantas98/tapioca](https://github.com/gabriel-dantas98/tapioca) - **License:** MIT - **Homepage:** https://gabriel-dantas98.github.io/tapioca 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:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **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-gabriel-dantas98-tapioca - Seller: https://agentstack.voostack.com/s/gabriel-dantas98 - 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%.