flompt

flompt is a visual AI prompt-building tool whose core idea is to break a long prompt down into structured semantic blocks, edit them in a flowchart interface, and then compile them into an XML-format prompt optimized for Claude. According to official materials, its slogan is "flow + prompt = flompt," that is, "turn any AI prompt into a visual flow: break it down, edit it into a flowchart, recompile it."

Purpose: flompt is aimed at prompt engineering scenarios, helping users stop writing a single long text and instead decompose prompts into typed structured blocks, then organize and output them as machine-readable XML according to the prompt structure recommended by Anthropic. It is suitable for users who need to repeatedly debug, reuse, or systematically manage prompts.

Main capabilities:

1. AI breakdown: after pasting any prompt, AI splits it into typed semantic blocks.
2. Visual editing: drag, connect, and reorder semantic blocks in a flowchart editor based on React Flow; provides switching between a canvas view and a card list view.
3. Compilation output: compiles semantic blocks into a Claude-optimized XML prompt and provides an estimated token count.
4. 16 specialized semantic blocks: directly corresponding to Claude prompt engineering best practices, and automatically ordered according to the structure recommended by Anthropic.
5. Block management: hide or show any block (hidden blocks do not participate in final prompt assembly), and copy any block with one click in the list view.
6. Asynchronous task queue: used for non-blocking breakdown and provides real-time progress.
7. Auto-save: through local Zustand persistence.
8. Export: supports export as TXT or JSON.
9. Multilingual: supports 10 languages in total: EN, FR, ES, DE, PT, JA, TR, ZH, AR, and RU, with a separate index page for each language.
10. Keyboard shortcuts.

How to use:

• Online use: visit flompt.dev, free, no account required.
• Browser extension: can be used in ChatGPT, Claude, and Gemini without leaving the current tab. The extension injects an Enhance button into the AI chat input box, with two-way synchronization between the sidebar and the chat, and supports Chrome and Firefox.
• Claude Code integration (MCP): flompt uses the Model Context Protocol to integrate its core capExposed as a native tool in Claude Code. Once configured, decompose_prompt, compile_prompt, and list_block_types can be called in any Claude Code session, with no browser or copy-paste required. Provided via CLI (recommended, --scope user makes the tools available in all Claude Code projects) or configured through ~/.claude.json.
• Three MCP tools: decompose_prompt breaks down a raw prompt into structured blocks (role, goal, context, constraints, etc.); if the server is configured with an API key, it uses Claude or GPT, otherwise it falls back to keyword-based heuristic analysis, returning a list of typed blocks and the full JSON for use by compile_prompt. compile_prompt compiles the block list into a Claude-optimized XML prompt, reorders it according to Anthropic's recommended structure, and returns the final XML along with an estimated token count. list_block_types lists all 16 block types along with their descriptions and recommended canonical order.
• Self-hosting: supports local development, requires a Node/build environment (the specific commands are not fully listed in the materials); AI configuration optionally uses an Anthropic or OpenAI API key, and when there is no API key it uses a keyword heuristic decomposer and can still compile structured XML. Production deployment uses the /projects/flompt directory, consists of Caddy and a FastAPI backend, both managed by supervisord, with a keepalive watchdog that checks every 30 seconds and automatically restarts on crashes; Caddy automatically manages TLS certificates via Let's Encrypt. The deployment script supports full deployment (build + restart + health check), build-only, restart-only, and other modes.

Use cases:

• Prompt engineering and prompt tuning.
• Organizing scattered prompts into structured, reusable XML prompts.
• Quickly enhancing prompts in the ChatGPT, Claude, and Gemini chat interfaces.
• Directly decomposing and compiling prompts via MCP in Claude Code workflows.
• Self-hosted deployment to meet privatization or customization needs.

Limitations and notes:

• The official materials do notprovide specific pricing plans, prices, or paywall information; they only state that it is free, open source, and requires no account.
• When there is no API key, AI decomposition uses a keyword heuristic fallback, and the materials do not explain in detail its effectiveness or capability boundaries.
• The browser extension explicitly supports ChatGPT, Claude, and Gemini; other platforms are not mentioned.
• The specific dependency versions, system environment, and other requirements for the self-hosting part are not fully listed in the materials.
• The project uses the MIT license, is built by Nyrok, and welcomes contributions of bug reports, features, translations, and documentation.