gut

io.github.KungiePublished by GitHub account Kungiev0.8.2Updated Oct 24, 2026

Judgment calls on small, cheap models: is this spam, which team, how urgent. YES, NO or UNSURE.

VerifiedSTDIODesktop onlyProductivity & WorkflowDeveloper ToolsAI & ML

Overview

AI-generated overview

A local MCP server that lets an assistant make cheap yes/no, category and rating judgments on many items using small models, with an explicit unsure option.

What you can do
Answer yes/no questions about items, such as whether a comment is spam or a task is done, with an unsure outcome when confidence is low. Assign items to one of a fixed set of categories, such as which team owns a ticket. Rate items on an ordered scale, such as urgency. Run the same judgment across many items in one pass, with caching and concurrency. Choose or cascade backends, from TypeSafe's Jev to local or OpenAI-compatible models.
When to use it
Useful when an assistant must triage or filter large batches of text, files, commits or search results where rules are too brittle and a frontier model is too slow or costly. Best for short, well-defined judgment questions rather than open-ended generation.
Requirements
Runs locally over stdio, typically launched with uvx from the gutfeel-mcp PyPI package. A backend must be configured via GUT_BACKEND; remote backends need TYPESAFE_API_KEY or OPENROUTER_API_KEY, and ollaya, ollama or openai backends need GUT_BASE_URL and GUT_MODEL. Local backends require the PyTorch extras.
Before you install
Sends the text being judged to whichever backend is configured, which may be a third-party service such as TypeSafe or OpenRouter; local backends avoid this. API keys are passed as environment variables. Judgments are probabilistic and can be wrong, so high-stakes decisions should route unsure results to a person. Costs are billed per input by remote backends.

Installation

In SourceWeft

  1. Open gut in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

[gut]

[PyPI] [CI] [Downloads] [Glama MCP server]

Judgment calls as one line of Python — built for Jev, and running on any small model.

Try it in your browser → The site runs gut's local model in the page: no key, no server.

Your code keeps running into questions that aren't logic: Is this comment spam? Which team owns this ticket? How urgent is it? Is the agent's task done? Until now there were three answers:

  • Regex and keyword rules — free and instant, and wrong the moment someone phrases it differently.
  • A frontier LLM — understands anything, at seconds and cents a call, with prose to parse.
  • Train a classifier — cheap to run, once you have the labelled data, the pipeline and the week.

There is a fourth: a small model made for exactly these questions. TypeSafe AI's Jev answers typed questions directly — a probability for yes, a distribution over options, a score on a scale — with nothing to generate or parse, billed on input only. gut is built around it, and makes it a line of code:

python
import gut
gut.configure(backend=gut.JevBackend())   # or just set TYPESAFE_API_KEY
if gut.likely(comment, "is spam"):    hide(comment)

No prompt, no parsing, no threshold — and no model named at the call site.

Three questions

python
gut.likely(ticket, "is a bug report")                      # yes / nogut.classify(ticket, Team)                                 # which one — an Enumgut.rate(ticket, ["can wait", "this week", "right now"])   # how much

It knows when it doesn't know

A regex never hesitates, and neither does an LLM. gut can:

python
match gut.likely(email, "the customer threatens to cancel", ask_human=True):    case gut.YES:    escalate(email)    case gut.NO:     auto_reply(email)    case gut.UNSURE: send_to_a_person(email)

Say how careful to be in words — lean="yes", stakes="high" — and gut works out the thresholds.

A thousand subjects, one line

python
spam = gut.each(comments).likely("is spam")     # one decision per comment, in orderteams = gut.each(tickets).classify(Team)

Jev gets concurrent requests, a local model batched passes, and nothing already cached is asked twice. @gut.semantic does the same for several questions about one subject.

Jev first, any model

Jev is the model gut is designed around. It is not the only one: the model is configuration, and the same line runs unchanged on any of these.

python
gut.configure(backend=gut.JevBackend())                            # TypeSafe AI's Jevgut.configure(backend=gut.JevBackend.openrouter())                 # Jev, through OpenRoutergut.configure(backend=gut.JevBackend.ollaya("winnow:e4b"))         # open decision model, Ollayagut.configure(backend=gut.ZeroShotBackend())                       # NLI model, on your CPUgut.configure(backend=gut.TransformersBackend("Qwen/Qwen3-0.6B"))  # small LLM, on your machinegut.configure(backend=gut.OpenAICompatibleBackend(                 # Ollama, vLLM, llama.cpp    "qwen2.5:1.5b", base_url="http://localhost:11434/v1"))gut.configure(backend=gut.OpenAICompatibleBackend("gpt-4.1-nano")) # OpenAIgut.configure(backend=gut.OpenAIDecisionsBackend())                # OpenAI's Decisions API

Or several at once. Cascade asks the cheapest model first and passes on only what it is unsure of:

python
gut.configure(backend=gut.Cascade(    gut.ZeroShotBackend(),   # free and local: settles the obvious    gut.JevBackend(),        # sees only what the first could not))

Every answer is a model's own probabilities, never parsed from text, and decision.model names the model that gave it. Your own model can be a backend too: here is how.

Install

bash
pip install "gutfeel[jev]"           # + JevBackend: TypeSafe, OpenRouter or Ollayapip install "gutfeel[local]"         # + ZeroShotBackend and TransformersBackend (PyTorch)pip install gutfeel                  # core: any OpenAI-compatible server; FakeBackend for testspip install gutfeel-mcp             # + the MCP server, gutfeel-mcp

The package on PyPI is gutfeel (gut was taken); the import is plain import gut. No model at hand? gut.FakeBackend(answers={"is spam": 0.97}) answers from fixtures, for tests.

From a shell, and for agents

The big model thinks; the small one decides, fast. An agent with gut judges a thousand files, commits or search results in one command instead of reading each one itself:

bash
git ls-files | gut filter "retries failed requests" --read-files --max-cost 0.50claude mcp add gut --env TYPESAFE_API_KEY=your-key -- uvx gutfeel-mcp   # MCP Registry: io.github.Kungie/gutnpx skills add Kungie/gut --skill gut     # teaches a coding agent when to reach for it

Docs

The documentation, one page per idea: Getting started · Backends · Knowing when it doesn't know · Asking everything at once · Async · Exact costs · Caching and observability · Command line · MCP server · Honest limitations. examples/ runs the same code on every backend.

Status and license

Pre-1.0, Apache-2.0. Every code block in these docs runs in the test suite · contributing

Source: README.md at commit 94bdc51

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

3
  1. v0.8.2LatestOct 13, 2026
  2. v0.8.1Oct 7, 2026
  3. v0.8.0Sep 29, 2026

Score

50% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts