Skip to content
botshub

Give Claude Structured Decisions

by manankapoor23

MCP serverOther

View on GitHub

Give Claude structured decisions: one label from your allowed set with explicit confidence, first-class abstention, and auditable evidence, plus a policy gate that decides when automation is allowed.

About Give Claude Structured Decisions

typed-decision is a Claude Code skill with Python helper scripts for getting structured decisions from Claude instead of prose. The README says Claude returns one label from a caller-defined set, a self-reported confidence, an abstain option and evidence tagged as observed, inferred or missing. Its framing is that Claude proposes and your code decides. The author calls it an experiment.

It includes a decide skill, a validator for shape and allowed labels, a gate script that applies thresholds and returns distinct verdicts, and a calibration script reporting metrics such as ECE and Brier score. Some adapters and a batch runner are listed as planned. It needs Python 3.10 or later with no third-party packages, plus Claude Code to use the skill. You copy the skill folder into a skills directory. It is MIT licensed and owned by manankapoor23. This summary is from the README.

Key features

  • Single-label decisions
  • Abstain option
  • Evidence tagging
  • Policy gate script
  • Calibration metrics

More other MCP servers

  • Mcp Proxy

    sparfenyuk

    Other

    A bridge between Streamable HTTP and stdio MCP transports

    2.8K GitHub starsPythonMIT
  • Mcp Remote

    punkpeye

    Other

    Connect an MCP Client that only supports local (stdio) servers to a Remote MCP Server.

    1.7K GitHub starsTypeScriptMIT
  • Kanzo

    emanuele-em

    Other

    Self-hosted, sharable kanban for humans and agents. One 4.5 MB binary, 4.5 MB of RAM, 10 to 60 tokens per agent reply, starts in 2 ms, zero setup.

    1.2K GitHub starsRustAGPL-3.0
  • Other

    į”Ÿæˆå’ŒįūŽåŒ–åŊįž–čū‘åŽŸį”Ÿ PPT įš„ Agent Skill

    484 GitHub starsPythonMIT
  • mcptoon

    activeing123

    Other

    MCP tool discovery burns 71,929 tokens. mcptoon drops it to 581 (−99.2%, measured). Zero dependencies.

    431 GitHub starsPythonApache-2.0
  • automcp

    NapthaAI

    Other

    Easily convert tool, agents and orchestrators from existing agent frameworks to MCP servers

    303 GitHub starsPythonApache-2.0
Suggest an improvement
Report this listing
Is this your project? Claim this listing