These add no external data — the value is the scaffold, which forces a model to externalise steps it would otherwise skip: state the counter-argument, name the hypothesis, score the option against the criterion, mark its own confidence. Because each is a separate package, the sensible pattern is to pick the one or two that fit the work in front of you; loading thirteen reasoning frames into one client mostly costs context.
A monorepo of separate MCP servers, each one imposing a named reasoning structure on a problem rather than fetching data from anywhere.
- Structured Argumentation — formal dialectical reasoning, claims and counter-claims held apart
- Scientific Method — hypothesis testing and evidence evaluation
- Decision Framework — structured decision analysis
- Collaborative Reasoning — the same problem worked from several perspectives
- Metacognitive Monitoring — knowledge assessment and confidence tracking
- Bias Detection — flagging potentially biased wording
- Constraint Solver — validating logical and numeric constraints
- Analogical Reasoning, Visual Reasoning, Ethical Reasoning, Narrative Planner, Goal Tracker and Multimodal Synthesizer round out the set
Nothing to authorize — no account, no key, no network calls to a vendor. Each server installs on its own; @waldzellai/structured-argumentation is the documented example, run with npx, and every server is also published as a Docker image.
One command — npm install @waldzellai/structured-argumentation