Occam
Discover parsimonious governing equations from noisy measurements via SINDy and PySR without leaving your agent session.
Overview
Occam is a Validate-phase MCP server that finds the simplest equations consistent with your data using SINDy and PySR symbolic regression.
What is this MCP server?
- Occam MCP exposes SINDy and PySR symbolic regression over a remote Fit endpoint
- Optimizes for simplest equations consistent with supplied data—not black-box overfitting
- Streamable HTTP at occam.fit for agent-driven science and ML prototyping
- Bridges spreadsheet or experiment exports to interpretable formulas your app can later codify
- Version 0.1.1 with published Model Context Protocol server schema
- Methods: SINDy and PySR via MCP
- Remote endpoint: https://occam.fit/mcp/ (streamable-http)
- Server version 0.1.1 per published MCP schema
What problem does it solve?
Builders drown in curve-fitting scripts and opaque models when they only need a readable law to decide if an idea is worth building.
Who is it for?
Solo builders validating physics-informed features, sensor dashboards, or scientific SaaS where explainable equations beat black boxes.
Skip if: Production forecasting that requires deep learning, vision, or NLP without interpretable structure.
What do I get? / Deliverables
After connecting Occam, your agent can return candidate symbolic formulas you can test, document, and optionally implement in your prototype.
- Candidate symbolic equations ranked for simplicity versus fit
- Agent-assistable regression runs without local PySR/SINDy environment setup
- Documentation-friendly formulas for prototype specs and investor demos
Recommended MCP Servers
Journey fit
Symbolic regression belongs in Validate because you are testing whether a simple law explains observed data before locking a full production model. Prototype is where solo builders run exploratory fits, compare candidate equations, and decide if the signal is real enough to implement in code.
How it compares
Symbolic-regression MCP integration, not a hosted notebook or a generic pandas skill.
Common Questions / FAQ
Who is Occam for?
Indie developers and researchers using coding agents who want SINDy- or PySR-style equation discovery from tabular or time-series measurements.
When should I use Occam?
Use it in Validate when prototyping models, benchmarking simple laws against baselines, or documenting governing equations before backend implementation.
How do I add Occam to my agent?
Add the remote MCP URL https://occam.fit/mcp/ (streamable-http) to your agent’s MCP configuration and pass datasets through the server’s regression tools.