just-prompt MCP
An MCP server that sends one prompt to several OpenAI, Anthropic, Gemini, Groq, DeepSeek and Ollama models at once and can draw a decision from them
- 739 ★ on the repository
- Last commit
Extended access
Why this level
- Sends prompt text and the content of files at an absolute path to external model providers and to a local Ollama
- Holds the keys of several paid APIs in the environment and spends money on every request, including model-name correction requests
- Writes answers as Markdown files into a folder whose path the agent passes
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/just-promptDetects 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.
- Sends prompt text and the content of files at an absolute path to external model providers and to a local Ollama
- Holds the keys of several paid APIs in the environment and spends money on every request, including model-name correction requests
- Writes answers as Markdown files into a folder whose path the agent passes
This is third-party code. Review the repository files before installing.
What it does
just-prompt is a small Python MCP server through which an agent talks to several language models from different providers at once. The `prompt` tool takes text and a list of models, runs the requests in parallel and returns every answer. `prompt_from_file` reads the prompt from a file at an absolute path, and `prompt_from_file_to_file` also saves each model's answer to its own Markdown file in a chosen folder. The `list_providers` and `list_models` tools show the available providers and their models. Models are written with a provider prefix, and each has a short form: `o:` for OpenAI, `a:` for Anthropic, `g:` for Gemini, `q:` for Groq, `d:` for DeepSeek, `l:` for Ollama. If a model name is slightly wrong, the server corrects it with a request to the first model in the `--default-models` list. At startup it checks which providers have keys in the environment and marks the rest unavailable while continuing with the others. A separate `ceo_and_board` tool sends a question to board-member models, passes their answers to a CEO model, `openai:o3` by default, and saves its decision with the answers as Markdown. The repository holds examples of such runs, including a choice between OpenAI, Anthropic and Google, and the `jprompt_ultra_diff_review` command for Claude Code: it saves `git diff` output to a file and sends it to several models for review.
Who it is for. For developers who use Claude Code and want to compare several models' answers to one question, review or architecture decision without switching between windows and services.
Good fit when
- You want to compare how OpenAI, Anthropic and Gemini answer the same prompt and save the answers as files with `prompt_from_file_to_file`
- Before a big stack or architecture choice you want several models' opinions and a verdict from one of them through `ceo_and_board`
- You want a git diff reviewed by several models at once with the `jprompt_ultra_diff_review` command
- Some models run locally through Ollama and some over an API, and you want one server for all of them
- You need a quick look at which models a provider offers, with `list_models`
Not a fit when
- You have no provider keys at all: without them the server returns nothing, and OpenAI and Anthropic APIs are unavailable from Russia
- You need streamed output or a dialogue with history: the tools work as one request and one answer
- You want a ready plugin with subcommands for Claude Code: this is a server installed from a clone with `uv`
- The code must have a formal license: the repository has no license file
Example request
Send this question through just-prompt to the models openai:gpt-4o and anthropic:claude-3-5-haiku and save both answers into the folder /Users/me/answersMCP
- Transport
- stdio
- Authentication
- key is optional
Environment variables
OPENAI_API_KEYsecret- OpenAI key, needed for models with the `openai` or `o` prefix
ANTHROPIC_API_KEYsecret- Anthropic key for models with the `anthropic` or `a` prefix
GEMINI_API_KEYsecret- Gemini key for models with the `gemini` or `g` prefix
GROQ_API_KEYsecret- Groq key for models with the `groq` or `q` prefix
DEEPSEEK_API_KEYsecret- DeepSeek key for models with the `deepseek` or `d` prefix
OLLAMA_HOST- Address of the local Ollama server if it is not the default
Limitations
- The author gave no license, so rights to use and copy are undefined; check with the author.
- You need your own provider keys in `.env` or the environment. You bear the cost of calls, and `ceo_and_board` hits several models at once and `openai:o3` as the CEO by default. OpenAI and Anthropic APIs are unavailable from Russia.
- The tools accept only absolute paths for files and folders, relative ones are not supported.
- The last commit in the repository dates from August 2025, and the model names in the examples and the default list belong to that time (for example `gemini-2.5-pro-preview-03-25`), so current names have to be substituted.
- The install instructions use a placeholder clone address with the name `yourusername`: the real repository is `disler/just-prompt`.
How to disable. Run `claude mcp remove just-prompt` to take the server out of Claude Code. In other clients delete the `just-prompt` entry from the MCP configuration. Delete the clone folder and the `.env` file with keys separately; model answers saved by the tools stay where you pointed them.
FAQ
Which providers does just-prompt support?
Six: OpenAI, Anthropic, Google Gemini, Groq, DeepSeek and Ollama. Each takes a key in `.env` (`OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, `GEMINI_API_KEY`, `GROQ_API_KEY`, `DEEPSEEK_API_KEY`) or an `OLLAMA_HOST` address for a local server.
How do I name a model in a request?
With a provider prefix and a colon: `openai:gpt-4o-mini` or shorter `o:gpt-4o-mini`. The other providers use the short prefixes `a`, `g`, `q`, `d`, `l`. If the prefix is omitted, the server uses the default models.
What does the ceo_and_board tool do?
It sends the question from a file to board-member models, passes their answers to a CEO model, `openai:o3` by default, and writes the CEO decision and the board answers to the `abs_output_dir` folder. The repository includes an example decision on choosing between OpenAI, Anthropic and Google.
Does just-prompt work without keys?
The server starts, but calls to providers without a key are unavailable. At startup it checks the environment and reports which providers can be used. Local Ollama models need no key, only a reachable `OLLAMA_HOST`.
Which language and version are required?
Python 3.10 or newer according to the `pyproject.toml` manifest. Dependencies (`anthropic`, `google-genai`, `groq`, `ollama`, `openai`, `mcp`, `pydantic`) are installed by `uv sync`. Tests run with `uv run pytest`.
What is the jprompt_ultra_diff_review command?
A Claude Code command from the `.claude/commands` folder. It creates a `diff.md` file with a review request, appends the `git diff` output and sends the file to three models through `prompts_from_file_to_file`. The models in the command are hard-coded as of its writing and are worth replacing.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server that lets Claude Code, Codex and other agents call Gemini, GPT, Grok and local models for review, debugging and consensus
An MCP server and web tool for improving prompts: three tools for optimizing system and user prompts with iterative refinement
A Claude Code plugin that asks several AI models the same question and shows their answers side by side with a synthesis of agreement and divergence
A CLI that turns any MCP server, OpenAPI spec or GraphQL endpoint into commands at runtime, with no codegen
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent
A CLI for any MCP server: list and call tools, resources and prompts, run mock and proxy modes, guard access and manage agent configs
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| just-prompt (this entry) | no | key is optional | 739 | |
| claude-council | no | 898 | a Claude Code plugin that queries models through API keys and subscriptions and shows answers side by side with an agreement analysis | |
| LLM Council | no | 2,672 | a skill where five sub-agents with different roles review each other, with no calls to other providers | |
| MCP CLI client | no | 675 | a console MCP client with one model chosen in a config |
README in short
The front page presents the server as a light bridge to six providers and shows the tool outline with parameters. Next come a table of provider prefixes, a feature list, install steps through `uv sync`, environment variables and three ways to connect the server to Claude Code. It ends with the code structure and how to run tests. The repository has sample answers, technical specifications, reference documents on provider APIs, layered tests and a ready `.mcp.json`.