CLI Reference

mtf "Describe your phenomenon here"
mtf "Describe your phenomenon here" --files plot1.png plot2.png
mtf   # interactive mode — you are prompted for everything

Flags

Flag	Default	Description
--n-literature	3	Parallel literature agents
--n-fitting	3	Parallel fitting agents per hypothesis
--n-qualitative	3	Parallel qualitative evaluation agents (used with --no-fitting)
--n-reviewer	3	Parallel reviewer agents
--n-proposal	2	Parallel measurement proposal agents
--max-debate-rounds	3	Max debate rounds before auto-proceeding
--literature-model	claude-opus-4-6	Model for literature agents
--fitting-model	claude-opus-4-6	Model for fitting agents
--reviewer-model	claude-opus-4-6	Model for reviewer agents
--proposal-model	claude-opus-4-6	Model for proposal agents
--debate-model	claude-opus-4-6	Model for debate synthesis
--image-digest-model	claude-opus-4-6	Model for image/document digestion
--files	(none)	Input files to digest (images: PNG, JPG, GIF, WebP; documents: PDF)
--physics-domains	condensed_matter	GPD physics domains (space-separated)
--no-fitting	(off)	Skip quantitative fitting; run qualitative hypothesis evaluation instead
--auto-detect-domains	(off)	Infer physics domains from the phenomenon text via GPD (opt-in)
--no-gpd	(off)	Disable GPD MCP physics verification
--gpd-servers	(all)	GPD servers to start
--no-enhanced-pdf	(off)	Use single-pass PDF extraction instead of the default two-pass figure analysis

Examples

# Quantum Hall phenomenon with two images
mtf "Plateau in rho_xx at B=3T" --files rho.png Hall.png

# Cross-domain (neutron star)
mtf "Neutron star cooling anomaly" --physics-domains gr nuclear amo

# Lightweight run — fewer agents, cheaper models
mtf "Raman peak shift" \
    --n-literature 2 --n-fitting 2 --n-reviewer 1 \
    --literature-model claude-haiku-4-5-20251001 \
    --fitting-model claude-haiku-4-5-20251001

# Disable GPD for faster iteration
mtf "Test phenomenon" --no-gpd