gut
Hands an agent's yes/no, category and rating questions to a small, cheap model
- Pricing
- Open source
- Platforms
- CLI, self-hosted
- Runs
- Locally
- Version
- 0.8.2
- Listing updated
- 2 days ago
About gut
gut is an open-source MCP server for ai models and llm tooling. It runs locally on your machine and is started by your AI client.
The premise is that a large agent model should not spend seconds and cents deciding whether a comment is spam. The gutfeel-mcp package passes simple judgment calls to a small model and returns a typed answer: an outcome with a probability, a chosen label with every option's probability, or a score on a scale. Answers are the model's own probabilities, never parsed from generated prose.
It runs locally over stdio and starts through uvx. The default model is Jev from TypeSafe AI when TYPESAFE_API_KEY is set. A GUT_BACKEND variable switches to other options such as OpenRouter, Ollama, an OpenAI-compatible server, OpenAI's Decisions API or a local zero-shot encoder, so a purely local setup can keep text on your machine. The model loads on first call, so the client handshake is instant.
The unsure outcome appears only when the agent passes ask_human: true, signalling that a person should read the text. The docs advise using it for simple, high-volume judgments, since for one subtle call the agent can read the text itself.
Install
Run in a terminal
claude mcp add gut --env TYPESAFE_API_KEY='<TYPESAFE_API_KEY>' --env OPENROUTER_API_KEY='<OPENROUTER_API_KEY>' -- uvx gutfeel-mcpAdd to claude_desktop_config.json
{
"mcpServers": {
"gut": {
"command": "uvx",
"args": [
"gutfeel-mcp"
],
"env": {
"TYPESAFE_API_KEY": "<TYPESAFE_API_KEY>",
"OPENROUTER_API_KEY": "<OPENROUTER_API_KEY>"
}
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"gut": {
"command": "uvx",
"args": [
"gutfeel-mcp"
],
"env": {
"TYPESAFE_API_KEY": "<TYPESAFE_API_KEY>",
"OPENROUTER_API_KEY": "<OPENROUTER_API_KEY>"
}
}
}
}Add to .vscode/mcp.json
{
"servers": {
"gut": {
"type": "stdio",
"command": "uvx",
"args": [
"gutfeel-mcp"
],
"env": {
"TYPESAFE_API_KEY": "<TYPESAFE_API_KEY>",
"OPENROUTER_API_KEY": "<OPENROUTER_API_KEY>"
}
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"gut": {
"command": "uvx",
"args": [
"gutfeel-mcp"
],
"env": {
"TYPESAFE_API_KEY": "<TYPESAFE_API_KEY>",
"OPENROUTER_API_KEY": "<OPENROUTER_API_KEY>"
}
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"gut": {
"command": "uvx",
"args": [
"gutfeel-mcp"
],
"env": {
"TYPESAFE_API_KEY": "<TYPESAFE_API_KEY>",
"OPENROUTER_API_KEY": "<OPENROUTER_API_KEY>"
}
}
}
}Configuration
TYPESAFE_API_KEYKey for TypeSafe's Jev, the model asked by default when it is setGUT_BACKENDAnother model: openrouter, ollaya, clm, ollama, openai, openai-decisions, zeroshot or transformersGUT_MODELThe model to ask; required for ollaya, ollama and openaiGUT_BASE_URLThe server's URL, for ollaya, ollama and openaiOPENROUTER_API_KEYKey for Jev through OpenRouter, with GUT_BACKEND=openrouter
Tools this server provides (4)
Taken from the project documentation; the running server may differ.
- likely
- Tests whether a claim is true of a text; returns yes, no or unsure and a probability
- classify
- Chooses which option fits and returns every option's probability
- rate
- Places text on a scale and returns a score and confidence
- each
- Asks the same question about up to 1000 texts
What it does
- Yes/no judgments
- The `likely` tool tests a claim against text and returns yes, no or unsure with a probability.
- Option classification
- `classify` picks a label from a list or a label-to-description map and reports every option's probability.
- Scale rating
- `rate` places text on a scale and returns a score, the nearest level and a confidence.
- Batch of up to 1000
- `each` asks the same question about up to 1000 texts and returns answers in order.
- Swappable backends
- One environment variable moves between Jev, OpenRouter, Ollama, OpenAI and local models.
- Cost reporting
- Answers include usage and cost when the backend reports a price.
Good for
- Flag spam comments before they reach a queue
- Route support tickets to the owning team
- Rate how urgent a ticket is
- Escalate emails that need a person when the model is unsure
- Label hundreds of texts in one call
Pricing
Open source
Package is free; the default Jev model needs a TypeSafe API key, with pricing not published here.
Prices change. Check gutpy.dev before you buy.
gut FAQ
Which model does it use?
Jev from TypeSafe AI by default when TYPESAFE_API_KEY is set; otherwise you choose a backend with GUT_BACKEND.
Can it run fully locally?
Yes. With GUT_BACKEND set to ollama or zeroshot, nothing leaves your machine, and a local model costs 0.
How do I install it?
Add it with uvx gutfeel-mcp in your client's MCP config; uvx fetches it on first use.
When does it return unsure?
Only when the agent passes ask_human: true and the model cannot tell.
Is gut free?
Yes. gut is open source, so you can use and self-host it for free.
How do I add gut to Claude Code?
Run this command in your terminal: claude mcp add gut --env TYPESAFE_API_KEY='<TYPESAFE_API_KEY>' --env OPENROUTER_API_KEY='<OPENROUTER_API_KEY>' -- uvx gutfeel-mcp
What are the best alternatives to gut?
The closest alternatives to gut are Claude Code Ultimate Guide, signafAIde, Toolkien, Lugh and HypePaper.
Alternatives to gut

Claude Code Ultimate Guide
MCP serverMCP server for searching the Claude Code Ultimate Guide from any MCP client
Runs locally- Open source
- 6.5K GitHub stars
- AI models and LLM tooling

signafAIde
MCP serverLocal MCP server to set up, verify and check coverage of signed agent action records
Runs locally- 5 tools
- Freemium
- Security

Toolkien
MCP serverOne MCP gateway for all your MCP servers, with code mode, permissions and audit
Not checked yet- 3 tools
- Remote
- Open source
- AI models and LLM tooling
Lugh
MCP serverWrite, check and dry-run Lugh agent programs with policy, budgets and approvals
Runs locally- 5 tools
- Open source
- AI models and LLM tooling
HypePaper
MCP serverResearch papers, benchmarks, citations and topic experts for AI assistants
- 58 tools
- Remote
- Freemium
- AI models and LLM tooling

One compact entry point that finds, describes and runs hundreds of pay-per-call data tools
- 3 tools
- Remote
- Freemium
- Developer tools