ZTL Judge
Judge logical claims and paradoxes with a deterministic core, not an LLM
- Platforms
- API, Web, self-hosted
- Runs
- Remotely
- Version
- 1.1.0
- Listing updated
- today
Similar MCP servers
About ZTL Judge
ZTL Judge is an MCP server for ai models and llm tooling. It runs as a hosted remote server, so there is nothing to install locally.
ZTL Judge is built on a zero-trust idea: a language model may translate a statement into a formal table, but it never decides whether the statement is true. The table language is called ZFL. Each row carries a fact, a status (T verified, F refuted, or Z unverified by default), a ground, and a plain-words meaning that a back-reading step uses to catch encodings that say the opposite of what you intended.
A deterministic core then returns a verdict that is always two-valued, plus a warranty grade and a passport listing unverified inputs. For self-referential systems it reports the grounded part and a quarantine set, with components labelled PARADOX, UNDERDETERMINED, INPUT or DOWNSTREAM.
The repository describes a local studio written in Python with the standard library only, run at localhost:8190, and 41 worked examples. The AI is optional: with no key you fill the table by hand. The registry lists a hosted Streamable HTTP endpoint for the MCP server, and covers claims, algebra derivation steps and sets defined by a property.
Install
Run in a terminal
claude mcp add --transport http ztl-judge https://api.ztljudge.com/mcpAdd to Settings → Connectors → Add custom connector
https://api.ztljudge.com/mcpAdd to ~/.cursor/mcp.json
{
"mcpServers": {
"ztl-judge": {
"url": "https://api.ztljudge.com/mcp"
}
}
}Add to .vscode/mcp.json
{
"servers": {
"ztl-judge": {
"type": "http",
"url": "https://api.ztljudge.com/mcp"
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"ztl-judge": {
"serverUrl": "https://api.ztljudge.com/mcp"
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"ztl-judge": {
"type": "http",
"url": "https://api.ztljudge.com/mcp"
}
}
}Using several servers? Build one config file with the MCP config generator.
What it does
- Deterministic verdicts
- A measured core judges; the LLM only translates.
- Warranty grades
- Verdicts carry hereditary, sound or until-verification grades.
- Paradox passports
- Self-referential systems get a quarantine set and per-component passports.
- Back-reading
- A non-AI step verbalises what the core read from the table.
- Hand-fill mode
- Works without any AI key.
- Worked examples
- Ships 41 examples to learn the table shape.
Good for
- Check a logical claim without trusting a model's judgement
- Analyse a self-referential paradox
- Audit whether an LLM's formalisation says what you meant
- Check algebra derivation steps
ZTL Judge FAQ
Does the AI decide the verdict?
No. The README says the LLM only translates the claim into ZFL, and the core judges.
Can I use it without an AI key?
Yes, in pro mode you fill the table by hand.
What does it need to run locally?
Python 3 with the standard library only; run python3 ztlstudio.py.
Does it handle arithmetic puzzles?
Not as propositional logic. The README says such wordplay gets an honest refusal instead of an invented encoding.
How do I add ZTL Judge to Claude Code?
Run this command in your terminal: claude mcp add --transport http ztl-judge https://api.ztljudge.com/mcp
What are the best alternatives to ZTL Judge?
The closest alternatives to ZTL Judge are REA, Context7, Chrome DevTools MCP, Codebase Memory and GitHub.
Where this comes from
- Description: written from the project's README and the project's website.
- Install steps: from the official MCP Registry entry.


