Skip to content

tuprolog/arg2p

v0.23.0MIT

Structured argumentation with Arg2P: formalize problems as theories, solve them, and explore the results in its solver.

GitHubOpen Website

Arg2P for AI assistants

Arg2P for AI assistants: structured argumentation (ASPIC+, ABA, abstract frameworks) solved by its engine, not guessed. The plugin bundles:

  • skills, the know-how of Arg2P:
    • arg2p-formalize: turning a problem with conflicting information, exceptions, norms, preferences or burdens of proof into a theory, with its profile and flags;
    • arg2p-solve: solving it, reading the labellings, checking them, reporting back;
    • arg2p-explore: showing the user the solver of the theory, to explore the results themselves;
  • the MCP server of Arg2P, which the skills solve with (its tools). It runs from the Python package of Arg2P with uv (uvx arg2p mcp), on the Java runtime the package bundles.

Where the client shows MCP Apps (Claude, ChatGPT, VS Code), explore shows the solver in the conversation: the solutions of the query, their argument graphs and stages, the drawing of an abstract framework, the flags. Everywhere, it links the web playground with the same theory.

Installing

The plugin needs uv for its MCP server. In Claude Code:

/plugin install arg2p --marketplace tuProlog/arg2p-kt

Claude (claude.ai, Claude Desktop), Codex and ChatGPT, the skills alone in any agent reading Agent Skills, and the MCP server alone in any MCP client: see AI assistants.