Truth values as frequency-and-confidence pairs are what separate this from asking a model to "consider multiple perspectives": positions accumulate evidential weight and get revised when something contradicts them. The author is candid that this is a research prototype — the semantic similarity is simple, and the synthesis step is not yet transparent enough for the model to reason about how a conclusion was reached.
A reasoning server that pairs a pluralist philosophical framework with NARS — a non-axiomatic reasoning system — via the pip-installed ONA engine. Beliefs carry frequency and confidence rather than being true or false, and contradictory evidence triggers revision instead of being discarded.
- `analyze_concept` examines a concept through multiple interpretive lenses, taking a context and a confidence threshold
- `explore_coherence` maps coherence patterns across a domain at a chosen depth, with revision allowed or not
- `contextualize_meaning` derives what an expression means inside a named language game and form of life
- `generate_insights` produces fallibilistic insights across a list of perspectives, with an openness coefficient
- `ask_philosophical_question` and `explore_philosophical_topic` are available when running the server directly
- Inference covers deduction, induction, abduction, analogy and revision, with uncertainty propagated along the chain
The Python package `openended-philosophy-mcp`, version 0.1.0, with the entry point `openended-philosophy-server`. Install with uv (`uv sync`), which pulls in `ona`. Launch through `uv --directory <path> run openended-philosophy-server`, setting `PYTHONPATH` and `LOG_LEVEL`. The NARS engine is tuned through `NARS_MEMORY_SIZE` (default 1000), `NARS_INFERENCE_STEPS` (default 50), `NARS_SILENT_MODE` (default true) and `NARS_DECISION_THRESHOLD` (default 0.6).