Warrant

You can run and interact with the warrant-mcp server to perform formal reasoning, argument validation, and structured dialogue for AI agents. This MCP server offers a suite of reasoning patterns from classic argumentation theories and exposes a set of tools that let you build, analyze, and compare arguments, manage dialogues, and diagnose disagreements in a principled way.

How to use

You run the MCP server locally and connect to it with an MCP client or integration. Start the server using the provided command, then issue tool calls or skill commands to perform argument construction, evaluation, dialogue, and disagreement diagnosis. You can use the built-in tools to build Toulmin-style arguments, identify Walton schemes, compare argument extensions under different semantics, score arguments with gradual semantics, run dialogue sessions, and diagnose disagreements between agents.

How to install

Prerequisites: you need a working runtime for the server. The setup described here uses the uv tool to manage dependencies and run the MCP server.

# Clone the repository
git clone https://github.com/jayden-chmod/warrant-mcp.git
cd warrant-mcp

# Install dependencies
uv sync

# Run the MCP server
uv run warrant-mcp

Configure the MCP server for a client integration by adding a configuration entry that points to the local server command. The following example shows how to reference the local server from a configuration file used by a client application.

{
  "mcpServers": {
    "warrant-mcp": {
      "command": "uv",
      "args": [
        "run",
        "warrant-mcp"
      ]
    }
  }
}

Additional content you should know

The server exposes a set of MCP tools and slash commands that let you perform structured reasoning workflows. You can start with building arguments, identifying argumentation schemes, evaluating defenses and defeaters, and running collaborative or adversarial dialogues. The system also provides mechanisms to score arguments, compare different framings, and diagnose where disagreements originate.

Available tools

build_argument

Build a structured Toulmin argument with claim, data, warrant, backing, rebuttal, and qualifier and return an argument, validation, and score.

identify_scheme

Identify which Walton argumentation scheme matches a claim or retrieve details for a specific scheme.

classify_defeater

Classify a counterargument as a rebutting or undercutting defeater and provide its strength and penalty.

create_framework

Create a Dung Abstract Argumentation Framework or a Bipolar AF using arguments, attacks, and optional supports.

compute_extensions

Compute acceptable arguments under Dung semantics such as grounded, preferred, stable, or all.

score_arguments

Score arguments on a continuous [0, 1] scale using gradual semantics with optional support relations.

create_dialogue

Start a Prakken-style dialogue session with a topic and participants and obtain the serialized dialogue state.

dialogue_move

Make a speech act move in an active dialogue session and update commitment stores.

diagnose_disagreement

Diagnose why two agents disagree and classify the root cause into factual, inferential, preferential, or goal-conflict categories.

list_schemes

List all available Walton argumentation schemes with their critical question counts.