Python API

Minimal example

import asyncio
from mtf import MTFConfig, MTFOrchestrator

report = asyncio.run(
    MTFOrchestrator(config=MTFConfig()).run(
        "Describe your phenomenon here"
    )
)
print(report)

With data and images

import asyncio, numpy as np
from mtf import MTFConfig, MTFOrchestrator
from mtf.toolkit.registry import ToolkitRegistry

# Register experimental data
toolkit = ToolkitRegistry()
toolkit.register_data("B_field", np.linspace(0, 10, 200))
toolkit.register_data("rho_xx", rho_xx_array)
toolkit.register_data("rho_xy", rho_xy_array)

# Optional: register a model function
def drude_model(B, n, mu):
    return 1 / (n * 1.6e-19 * mu) / (1 + (mu * B) ** 2)

toolkit.register_model("drude", drude_model)

# Configure
config = MTFConfig(
    n_literature=3,
    n_fitting=3,
    n_reviewer=2,
    max_debate_rounds=2,
    physics_domains=["condensed_matter"],
)

# Run
orchestrator = MTFOrchestrator(config=config, toolkit=toolkit)
report = asyncio.run(
    orchestrator.run(
        "Anomalous resistivity plateau at B=3T in 2DEG at T=4K",
        images=["rho_vs_B.png", "Hall_plot.png"],
    )
)

MTFConfig parameters

Parameter	Default	Description
n_literature	3	Parallel literature agents
n_fitting	3	Parallel fitting agents per hypothesis
n_reviewer	3	Parallel reviewer agents
max_debate_rounds	3	Max synthesis iterations per phase
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
debate_model	claude-opus-4-6	Model for debate synthesis
image_digest_model	claude-opus-4-6	Model for image digestion
fitting_semaphore_limit	6	Concurrent fitting agent cap
physics_domains	["condensed_matter"]	GPD domain list
enable_gpd_mcp	True	Enable GPD physics verification