English Français العربية

gutsy

Registre officiel IA et agents

Local gut check before risky agent actions: calibrated safe/intended probabilities, no API calls.

Code source Documentation

Installation

Ajoutez ceci à la configuration MCP de votre client (Claude Desktop : claude_desktop_config.json ; Cursor : .cursor/mcp.json ; VS Code et les autres utilisent la même forme « mcpServers »).

PYPI package gutsy-mcp
Claude Code 
claude mcp add gutsy -e 'GUTSY_URL=http://127.0.0.1:8765' -e 'GUTSY_INFERENCE_DIR=<GUTSY_INFERENCE_DIR>' -e 'GUTSY_PROCEED_MIN=0.85' -e 'GUTSY_BLOCK_MIN=0.50' -- uvx gutsy-mcp==0.1.2
Configuration JSON (Claude Desktop, Cursor, VS Code) 
{
    "mcpServers": {
        "gutsy": {
            "command": "uvx",
            "args": [
                "gutsy-mcp==0.1.2"
            ],
            "env": {
                "GUTSY_URL": "http://127.0.0.1:8765",
                "GUTSY_INFERENCE_DIR": "<GUTSY_INFERENCE_DIR>",
                "GUTSY_PROCEED_MIN": "0.85",
                "GUTSY_BLOCK_MIN": "0.50"
            }
        }
    }
}

Variables d'environnement

  • GUTSY_URL URL of the local gutsy-inference server
  • GUTSY_INFERENCE_DIR Directory with models.json; if set, gutsy-mcp starts gutsy-inference itself
  • GUTSY_PROCEED_MIN Minimum P(proceed) and P(intended) to recommend proceeding
  • GUTSY_BLOCK_MIN Minimum P(block) to recommend blocking

Documentation

Tirée du README du projet sur GitHub, actualisée chaque semaine. Les liens ouvrent GitHub.

Sommaire
  1. gutsy-mcp
  2. What it is and isn't
  3. 1. Start the gutsy runtime
  4. 2. Add the MCP server
  5. 3. Optional: enforce it with a Claude Code hook
  6. Tool

gutsy-mcp

A local gut check for coding agents. Before Claude Code, Codex, Cursor or Copilot runs rm -rf, force-pushes, deploys or sends something, it calls gut_check and gets back calibrated probabilities that the action is intended and safe, plus a recommendation: proceed, confirm (ask the user first) or block.

Runs on your CPU with gutsy (0.8B, 775 MB GGUF). No API calls, no per-call cost, deterministic, and your code never leaves the machine.

What it is and isn't

It's a fast second opinion that adds friction where it's warranted. It is not a security boundary: the agent decides whether to call it and writes the context it sees, and a 0.8B model is weak at ambiguity and multi-step rules (see the model card). Keep your normal permission settings on. Default thresholds aren't tuned for your workflow; log the probabilities and adjust them.

1. Start the gutsy runtime

git clone https://github.com/kouhxp/gutsy
pip install -e gutsy/gutsy-inference
hf download kouhxp/gutsy --include "gutsy-0.8b-v04*" --local-dir gutsy/gutsy-inference/models
cd gutsy/gutsy-inference && cp models.example.json models.json
gutsy-inference check && gutsy-inference serve     # http://127.0.0.1:8765

Or set GUTSY_INFERENCE_DIR=/path/to/gutsy/gutsy-inference and gutsy-mcp starts it on first use.

2. Add the MCP server

Claude Code:

claude mcp add gutsy -- uvx gutsy-mcp

Cursor (.cursor/mcp.json) / Claude Desktop:

{ "mcpServers": { "gutsy": { "command": "uvx", "args": ["gutsy-mcp"] } } }

VS Code (.vscode/mcp.json):

{ "servers": { "gutsy": { "type": "stdio", "command": "uvx", "args": ["gutsy-mcp"] } } }

Codex (~/.codex/config.toml):

[mcp_servers.gutsy]
command = "uvx"
args = ["gutsy-mcp"]

Then tell the agent when to use it (CLAUDE.md, AGENTS.md, .cursor/rules):

Before deleting files, rewriting git history, force-pushing, touching databases, deploying, publishing or sending anything, call gut_check. If it doesn't return proceed, stop and ask me.

3. Optional: enforce it with a Claude Code hook

An MCP tool only runs if the agent remembers to call it. The hook runs on every risky-looking Bash command regardless. It can only ask or deny, never auto-approve.

uv tool install gutsy-mcp      # puts gutsy-hook on PATH

.claude/settings.json:

{
  "hooks": {
    "PreToolUse": [
      { "matcher": "Bash", "hooks": [{ "type": "command", "command": "gutsy-hook", "timeout": 60 }] }
    ]
  }
}

block becomes a confirmation prompt by default; set GUTSY_HOOK_ALLOW_DENY=1 to deny outright.

Tool

gut_check(action, context="", user_request="") returns:

{
  "recommendation": "confirm",
  "reason": "Not confident enough to proceed without the user.",
  "p_intended": 0.41,
  "p_reversible": 0.08,
  "verdict_probabilities": {"proceed": 0.22, "confirm": 0.61, "block": 0.17},
  "reject": 0.04,
  "confidence": 0.42
}
env vardefault
GUTSY_URLhttp://127.0.0.1:8765runtime URL
GUTSY_INFERENCE_DIRlets gutsy-mcp start the runtime
GUTSY_MODELruntime defaultmodel name from models.json
GUTSY_PROCEED_MIN0.85
GUTSY_BLOCK_MIN0.50
GUTSY_API_KEYif you run serve --api-key-env

License: Apache 2.0.

Questions fréquentes

Qu'est-ce que gutsy ?

Local gut check before risky agent actions: calibrated safe/intended probabilities, no API calls.

Comment installer gutsy ?

Il s'exécute sur votre machine. Dans Claude Code : claude mcp add gutsy -e 'GUTSY_URL=http://127.0.0.1:8765' -e 'GUTSY_INFERENCE_DIR=<GUTSY_INFERENCE_DIR>' -e 'GUTSY_PROCEED_MIN=0.85' -e 'GUTSY_BLOCK_MIN=0.50' -- uvx gutsy-mcp==0.1.2. Pour les autres clients, collez la configuration JSON de cette page dans leurs réglages MCP.

Faut-il une clé API pour gutsy ?

Son entrée au registre n'indique aucune clé ni secret obligatoire. Certaines fonctions peuvent tout de même demander un compte auprès du service auquel il se connecte.

gutsy est-il open source ?

Oui. Son code est sur GitHub sous licence Apache-2.0.

Qui maintient gutsy ?

Il est publié par kouhxp. AI.tn le référence sans le maintenir ni le recommander.

Issu du registre MCP officiel. Être listé n'est pas une recommandation : vérifiez le code et les permissions d'un serveur avant de lui donner accès à vos données.