flompt MCP
An MCP server and editor: a prompt is split into blocks, edited as a flowchart and compiled into XML for Claude
- 155 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Prompt text goes to the remote flompt.dev server, where an external model may process it if a key is set
- The browser extension injects into ChatGPT, Claude and Gemini pages
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/flomptDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Prompt text goes to the remote flompt.dev server, where an external model may process it if a key is set
- The browser extension injects into ChatGPT, Claude and Gemini pages
This is third-party code. Review the repository files before installing.
What it does
flompt splits a long prompt into typed blocks that you can move around a flowchart and compile back into machine-readable text. There are 16 block types: role, objective, context, constraints, guardrails, examples, chain of thought, output format, language and more. The compiled prompt is XML, with blocks ordered by Anthropic's recommendations. In Claude Code flompt connects as the remote MCP server `https://flompt.dev/mcp/` with no key and exposes three tools. `decompose_prompt` splits a raw prompt into blocks and returns JSON. `compile_prompt` compiles blocks into XML with an estimated token count. `list_block_types` lists the block types and the recommended order. The same features exist without an agent: a web editor at flompt.dev, free and without an account, and a Chrome and Firefox extension that adds an Enhance button to the ChatGPT, Claude and Gemini input boxes.
Who it is for. People who write long system prompts for Claude and want to see their structure in addition to the raw text.
Good fit when
- A page-long system prompt has to be split into role, constraints and examples
- A finished prompt must be compiled into XML tags that Claude reads better than a wall of text
- In Claude Code you want to decompose a prompt right in the conversation without opening a browser
- Moving prompt blocks with a mouse on a flowchart beats editing paragraphs
- ChatGPT, Claude or Gemini needs a prompt-improvement button in the chat window itself
Not a fit when
- The prompt holds trade secrets or personal data: decomposition runs on the flompt.dev server
- You want to score model output quality: flompt builds the prompt but does not test the result
- For one-line requests, splitting into blocks is not needed
Example request
Decompose this prompt into blocks with flompt and compile XML for Claude, adding a separate block for output format constraintsMCP
- Transport
- http
- Authentication
- not required
Limitations
- If the server has an Anthropic or OpenAI key, a model does the decomposition; otherwise a keyword heuristic runs. The README does not say which mode flompt.dev uses.
- The transport is Streamable HTTP, the server is stateless and every call is independent.
- DNS rebinding protection is on, and the README explicitly allows only the flompt.dev host.
- The last commit in the repository dates from March 2026.
- Self-hosting needs Python 3.12+ and Node.js 18+, and a key is optional.
How to disable. `claude mcp remove flompt` removes the server from Claude Code, and if you configured it in `~/.claude.json` delete the flompt block under `mcpServers`. The extension is disabled in the browser's extension manager.
FAQ
How do I connect flompt to Claude Code?
Run `claude mcp add --transport http --scope user flompt https://flompt.dev/mcp/`. The `--scope user` flag makes the server available in all projects. No key is needed.
Which tools does the flompt MCP server provide?
Three: `decompose_prompt` splits a prompt into blocks, `compile_prompt` compiles blocks into XML with a token estimate, and `list_block_types` shows the 16 block types and their order. The output of the first feeds the second.
Can I run flompt myself?
Yes. The README describes starting the FastAPI backend with `uvicorn app.main:app --reload --port 8000` and the frontend with `npm run dev`. An Anthropic or OpenAI key in `.env` is optional; without it the heuristic runs.
Does flompt work without a key?
Yes. The MCP server needs no authorization, and the web editor is free without an account. An Anthropic or OpenAI key on the server is needed only for model decomposition; without it a keyword heuristic runs.
Which blocks does flompt build?
16 types, including Role, Objective, Context, Constraints, Guardrails, Examples, Chain of Thought, Output Format and Language. Each maps to an XML tag, and the order follows Anthropic's recommendation.
Discussion
No questions or reviews yet.
Related
MCP serversMCP server for prompt templates, step chains and output validation, exported as skills for Claude Code, Cursor and Codex
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent
An experimental set of structured prompts for Claude: XML markup, the think tag and several variants in Lisp and TM formats. No license
A prompt template store with search and variable filling, plus a script that runs Claude agents across a whole project.
An MCP server and web tool for improving prompts: three tools for optimizing system and user prompts with iterative refinement
The author's personal Claude Code toolbox: meta-prompt and todo commands, skills to create skills, hooks and MCP servers, auditors and a Ralph loop
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| flompt (this card) | no | not required | 155 | |
| Prompt Optimizer | yes | not required | 41,320 | an MCP server and web tool that improves prompts through optimization, without a block editor |
| Prompt Architect | no | 333 | a skill that scores a prompt against criteria and picks a framework, with no block flowchart |
README in short
The README opens with a live demo and the idea, then lists the 16 block types with their XML tags. After that come the browser extension, an MCP section with installation, the three tools and a table of technical properties, and finally self-hosting instructions.